Stock

Romina Boccia and Dominik Lett

If you are congressional staff and you haven’t registered yet: Tomorrow, May 3, from noon to 1:15 P.M. EST, I will be joined by select fiscal experts for lunch and a deep dive into emergency spending. We will discuss how the abuse of emergency spending has worsened the already grave fiscal outlook and explore innovative proposals for budget reform. Hope to see you there!

Register here. 

Congress is increasingly abusing emergency designations, which were originally intended to provide a safety valve for responding to urgent and unexpected crises, to evade spending limits. Over the last three decades, one in every ten dollars of federal spending has been exempted from normal budget controls through emergency‐​related designations. This fact sheet aims to provide key fiscal details that legislators and the public should know about the growing use of emergency designations.

Congress provided $12 trillion in spending for emergencies over the past 30 years.

Discretionary budget authority accounts for half of emergency‐​related spending. Defense and disaster relief regularly receive “emergency” funding for recurring, predictable operations.

Mandatory or direct budgetary authority accounts for the other half of emergency spending, with 87 percent of this spending traceable to just five pandemic and Great Recession laws.

Emergency spending comprised 3 percent of annual budget authority following the 9/11 terrorist attacks. It grew to 18 percent following the Great Recession. Emergency spending peaked at 32 percent of annual budget authority in 2020—an inflation‐​adjusted $3 trillion (see figure below).

Because most emergency spending is deficit spending, we estimate emergency spending generated an additional $2 trillion in interest costs. That brings the grand emergency total to $14 trillion—roughly half of the size of the government’s $27.5 trillion in public debt.

Congress continued emergency spending even after the pandemic ended.

In 2023, Congress provided $162 billion in emergency designations. This funding was primarily allocated for Ukraine aid, disaster relief, and prior‐​year legislation, including the Infrastructure Investment and Jobs Act.

In 2024, Congress continued that trend with annual emergency spending now at $196 billion.

Together, three recent supplementals for Ukraine, Israel, and the Indo‐​Pacific region provided $96 billion in budget authority. This spending will generate $40 billion in additional interest costs over the next ten years.

To date, Congress has provided $174 billion in emergency funds for Ukraine. Including interest costs pushes the total above $240 billion over the next ten years.

Like in prior years, Congress used emergency funds in the 2024 yearly appropriations bills to evade spending limits. For example, more than half a billion dollars in law enforcement salaries were emergency designated.

Myopic abuse of emergency designations contributes to the long‐​run fiscal challenge.

During emergencies, Congress overspends because it is not bound by budget rules.

Unlike World War II, recent emergencies have resulted in sustained, elevated debt. This is largely because legislators are unwilling to make the necessary spending cuts to achieve primary surpluses. The nation’s fiscal projections are already dismal, and that’s before considering the possibility of a future conflict, pandemic, or some other crisis.

Congress is also prone to spend wastefully because of excessive emergency spending.

The Government Accountability Office found that emergency supplemental spending faced less oversight and that issues unrelated to the initial emergency often received funding.

The Associated Press found that $403 billion of COVID-19 relief funding, or 1 in every 10 dollars of pandemic aid disbursed, was stolen, wasted, or misspent.

Emergency Spending Fact Sheet

Andrew Gillen

Note: This post updates last month’s post. The biggest changes from last month include:

The newest plan relying on regulatory changes under the Higher Education Act has been released and is summarized.
A new court case against the SAVE plan has been noted with links to optimistic and pessimistic commentary on the chances of overturning the plan.
Updated the legal status of the borrower defense and closed school discharge plans to reflect an injunction from the Fifth Circuit.

Mass student loan forgiveness is a terrible policy (see this report for a comprehensive list of reasons), but that hasn’t stopped the Biden administration from trying to forge ahead. While the Supreme Court overturned the administration’s student loan forgiveness plan, every few weeks the executive branch announces another batch of loans that have been forgiven. The administration recently celebrated that since taking office it has succeeded in forgiving $143.6 billion of student loans for around four million borrowers by transferring the financial burden from the students who took out the loans to taxpayers who did not. And they aren’t going to stop—the administration declared that “President Biden has vowed to use every tool available to cancel student debt for as many borrowers as possible, as quickly as possible.”

But if student loan forgiveness lost in the Supreme Court, how are so many student loans still being forgiven? The answer is that there isn’t a student loan forgiveness plan. There are many plans, some of which are already up and running. Previous laws had already left a plethora of methods to forgive student loans, and many of those laws give the Secretary of Education the ability to expand those programs.

The administration is also claiming that existing law gives it the authority to create new ways to forgive student loans. So the student loans the Biden administration already has or wants to forgive are a combination of existing programs, programs the administration has expanded, and new programs the administration is seeking to implement.

Here’s a rundown of the administration’s student loan forgiveness plans and actions, which I’ll update monthly.

HEROES (New plan—overturned in court)

This was the big plan that got a lot of attention in 2022 and 2023. The plan was to forgive $10,000 for borrowers making less than $125,000, and $20,000 for borrowers who received a Pell Grant, at a total cost of $469 billion to $519 billion. The alleged authority for the plan was the 2003 HEROES Act. While designed to alleviate loan‐​related hardships for soldiers and reservists serving in Iraq and Afghanistan, the law also covered national emergencies, and the Biden administration argued the COVID-19 emergency gave it the authority to forgive virtually everyone’s loans. Most observers were skeptical of this supposed authority, but it was not clear who had standing to sue (standing is the requirement that those filing the suit have a concrete injury from the policy). The companies that service student loans would be the most obvious injured party, but there was a perception that the Biden administration would punish any servicer that challenged the policy in court, a perception that now appears accurate.

Fortunately, the Supreme Court ruled that Missouri had standing to sue (due to a quasi‐​public student loan servicer that would lose revenue under the plan), and that the plan violated the major questions doctrine (which holds that there needs to be clear congressional authorization for programs of substantial economic or political significance), preventing the policy from being implemented.

Higher Education Act (New plan—forthcoming)

Immediately after losing on HEROES, the Biden administration announced a new effort that would use authority under the Higher Education Act. The administration announced the new plan, which would

Waive unpaid interest.
Forgive debt for those who have repaid for 20 years (25 years if there is debt for graduate school).
Forgive debt for those who attended a low‐​financial value program (e.g., programs or colleges that fail the Cohort Default Rate or Gainful Employment).
Release additional regulations soon (under a pending plan) that will forgive debt for those undergoing economic hardship.

There are several problems with this plan, which the Penn Wharton Budget Model estimates will cost $84 billion. The public has a few more weeks to comment on the proposed regulations, and the administration will then consider the comments and issue final regulations, with a goal to start forgiving debt this fall. Once finalized, this plan will likely face the same fate in court as the HEROES plan, since it too will likely run afoul of the major questions doctrine. However, much of this forgiveness is easy to implement, so a key question is whether a court injunction will come fast enough to prevent the administration from forgiving billions of debt before the courts can determine whether the regulations are legal.

SAVE (New plan—still active)

Before diving into this one, it is important to understand the concept of income‐​driven repayment (IDR). Under traditional (mortgage) style loan repayment, the amount and length of repayment are fixed (e.g., $200 a month for 10 years). For the past few decades, the federal government has been introducing IDR plans, in which the amount repaid each month varies based on the borrower’s current income and the length of repayment varies based on how fast they repay their loan. The key features of an income‐​driven repayment plan are:

the share of income owed each month (e.g., 20%);
the income exemption that is protected from any repayment obligation (e.g., the poverty line);
the cap on length of repayment (e.g., 25 years).

IDR is a great idea, providing borrowers with better consumption‐​smoothing across their lifetime and flexible repayment, which helps avoid defaults due to short‐​term liquidity constraints.

But we’ve also botched the implementation. To begin with, a cap on the length of repayment is completely inappropriate. Income‐​driven repayment ensures that payments are always affordable, and borrowers who make so little they do not repay will receive de facto forgiveness even without the cap, so there is no justification for a cap.

The other problem with how we’ve implemented income‐​driven repayment is political—the plans are tailor‐​made to allow politicians to give constituents big benefits today while sticking future taxpayers with the bill. It is, therefore, no surprise that these plans have grown more generous over time. The first IDR plan, introduced in 1994, had an income exemption equal to the poverty line, a share of income owed of 20 percent, and a cap on length of twenty‐​five years. Very few borrowers would receive forgiveness under these terms and, of those who did, they really wouldn’t have been able to repay regardless of whether they received forgiveness. The Obama administration introduced plans with an income exemption of 150 percent of the poverty line, a share of income owed of 10 percent, and a cap on length of payment of twenty years.

The Biden administration’s Saving on a Valuable Education (SAVE) plan took an existing plan (the REPAYE plan) and made it much more generous. It changes the share of income owed from 10 percent to 5 percent, increases the income exemption from 150 percent of the poverty line to 225 percent, and caps the length of repayment at as little as ten years for some borrowers.

By cranking every possible lever to the most generous settings in history, this plan would impose massive costs on taxpayers, estimated at $475 billion for just the next ten years.

Parts of the SAVE plan have already been implemented, and full implementation is scheduled for July 2024. However, there are now two lawsuits that seek to overturn the plan, one by Kansas and ten other states, and another by Missouri and six other states. The legal questions facing this plan are the reverse of the HEROES plan. For the HEROES plan, the main obstacle was standing. Once that hurdle was cleared, it was fairly obvious that the plan was well beyond what Congress had authorized. But for the SAVE lawsuits, this is reversed. Standing is easily established (for Missouri at least), but the plan does have a much stronger argument for being within the parameters of the law. Mark Kantrowitz thinks SAVE will be upheld, while Michael Brickman did yeoman’s work digging up details on page 18,909 of the 1993 Congressional Record that may lead to SAVE being scrapped.

Student Loan Payment Pause (Existing and extended plan—now expired)

When COVID-19 hit in March 2020, student loan payments were paused. The pause was supposed to last two months but ended up lasting three‐​and‐​a‐​half years after Trump extended it once and Biden extended it six times. A pause would not normally result in massive student loan forgiveness as it would delay, but not waive, repayment. There would still be a cost to taxpayers (driven by the government’s cost of borrowing), but it wouldn’t be huge. But recall that IDR plans (unnecessarily) cap the length of repayment, and the pause counted towards that cap.

In other words, for any student who does not fully repay before they hit the length of repayment cap, payments weren’t paused, they were waived. We won’t know for many years how many students had their payments forgiven rather than postponed, but the current estimates range from $210 billion to $240 billion.

There is virtually no chance for this burden on the taxpayer to be reversed. The only good news is that the payment pause ended, with most borrowers restarting payments in October 2023.

Public Service Loan Forgiveness (Existing and extended plan—still active)

The Public Service Loan Forgiveness (PSLF) program was established during the George W. Bush administration and allowed for public and nonprofit workers to receive forgiveness after ten years of repayment when they used an IDR plan. While I object to PSLF in principle (as a distorting and non‐​transparent subsidy for the government and nonprofit sectors) and due to the windfalls these borrowers receive (an average of over $70,000 per beneficiary), since PSLF legally exists, it should operate as seamlessly as possible. The Biden administration granted many waivers and other changes to increase the number of borrowers who could benefit under PSLF. Some of these changes were good in the sense that they more faithfully implemented the law, but the administration crossed some lines too. In particular, it started counting some types of deferment as payments (borrowers can get deferment when they cannot afford to make payments, which generally allows the borrower to temporarily postpone payments though interest continues to accrue). The whole point of deferment is to temporarily avoid making payments. So for the Biden administration to give borrowers credit for making payments when they were in deferment is logically, morally, and potentially legally wrong (Cato is part of a lawsuit seeking to end this abuse). The administration also waived income requirements, making more people eligible for the program.

The Biden administration has forgiven “$62.5 billion for 871,000 borrowers since October 2021” under these programs, which works out to just under $72,000 per borrower. By comparison, a formerly homeless student who receives the maximum Pell Grant for four years would get less than $30,000 in Pell Grants. Some of this would have been forgiven even if the administration hadn’t made any changes to the program, but not all of it. In the future, these burdens on the taxpayer can be reduced by rolling back some of the administrative changes, but eliminating the program entirely would require legislation.

Borrower defense to repayment (Existing and extended plan—still active, though recent changes are paused during a court case)

When a college engages in fraud or severely misleads students, borrowers can have their debt forgiven under borrower defense to repayment. This is reasonable, as victims of fraud should have some recourse. It is also extremely rare because a college would not just need to dupe a student but would also need to fool a state, an accreditor, and the US Department of Education, as all three are required to sign off on the legitimacy of a college before its students can take out student loans. As the House Committee on Education & the Workforce noted, “for the first 20 years of the rule, there were 59 claims.”

However, the federal government can claw back debt forgiven from the responsible college. This makes borrower defense to repayment an incredibly powerful tool for progressives in their war on for‐​profit colleges. If a for‐​profit college can be declared to have substantially misled students, they can be ruined financially by the clawbacks. Indeed, new regulations from the Biden administration would make it much easier to conclude a college engaged in misconduct. As the White House gloated, “Less than $600 million in debt relief had been approved through borrower defense, closed school discharges, and related court settlements from all prior administrations combined, compared to the $22.5 billion approved under the Biden‐​Harris Administration alone.” Some of this was done outside the law. For example, $5.8 billion of debt for Corinthian College students was forgiven even if students didn’t submit a borrower defense claim. The administration has promised to forgo clawbacks on much of it (likely in part to avoid giving affected colleges standing to oppose the changes in court).

The good news is that any further forgiveness under the new regulation is on hold due to an injunction from the Fifth Circuit Court of Appeals (this injunction applies to the closed school discharge plan as well).

Closed School Discharge (Existing and extended plan—still active, though recent changes are paused during a court case)

Borrowers whose school closes while they are still enrolled or shortly after they have withdrawn can have their student loans forgiven. The Biden administration imposed new regulations that loosened the requirements and has used this as an excuse to forgive other loans as well. For example, Biden forgave $1.5 billion in debt for students from ITT Technical Institute, even if they didn’t qualify for a discharge. Further forgiveness under the new regulations has been paused by the Fifth Circuit Court of Appeals until courts determine whether the new regulations are legal. However, the administration can still forgive loans under the previous iteration of these regulations. In fact, the administration just announced another $6.1 billion of forgiveness today (May 1).

Total and Permanent Disability Discharge (Existing and extended plan—active)

Borrowers who are unable to work due to a permanent disability can have their loans forgiven. Historically this was very rare. To protect against fraud, the income of borrowers who had their debt forgiven was monitored to ensure that they really couldn’t work. The Biden administration both expanded eligibility and dropped fraud detection efforts, leading forgiveness under total and permanent disability discharge to spike from negligible amounts to $11.7 billion.

Conclusion

In sum, the Biden White House has been the most aggressive presidential administration in history regarding student loan forgiveness. Despite many setbacks, the administration has canceled a massive amount of debt, with most of the burden on taxpayers still to come from future repayments that will no longer be made. And while many of its attempts to forgive student loans have been stymied, there are still many active plans in play, with more on the horizon.

Walter Olson

Libertarian thinkers have long proposed that courts should do more to protect the right to earn a living, more precisely articulated as the right to pursue the occupation of one’s choice as against government‐​granted monopolies and arbitrary curbs on entry to a chosen trade. Timothy Sandefur has noted that this right “was protected under English common law as far back as the Elizabethan era,” and was seen by both proponents and opponents as part of the set of rights protected by the Fourteenth Amendment. As a fundamental right, Sandefur writes, the freedom to pursue an occupation “should receive the protections of review under the strict scrutiny standard, rather than the deferential rational basis test.” Yet as we know, the federal courts chose a different path.

You might be surprised to learn that North Carolina, unique among the states, includes in its constitution a clause that covers much of this ground. Adopted in 1868 after the Civil War, the document proclaims in Article I, Section 1 (boldface added):

We hold it to be self‐​evident that all persons are created equal; that they are endowed by their Creator with certain inalienable rights; that among these are life, liberty, the enjoyment of the fruits of their own labor, and the pursuit of happiness.

In the Twentieth Century, amid hesitations and false starts, the Tarheel State’s courts began to treat this language as a serious source of enforceable rights. Cases as far back as the 1940s applied the clause to strike down arbitrary licensing restrictions on photographers and dry cleaners. In later cases from Wilmington and Durham, they held that the clause could protect government employees who lose their jobs if the cities that employ them had not followed proper legal rules. (I won’t try to address here whether “on the public payroll” should be an implied dimension of the stated right.)

Last month, the state’s intermediate appellate court, the North Carolina Court of Appeals, ruled that bar owners could use the clause (as well as the state’s equal protection clause) to challenge as arbitrary a 2020 order by Governor Roy Cooper closing some but not other bars on grounds of risk of Covid contagion. A 2022 decision from the same court applied similar reasoning in a case brought by a racetrack that had held events with social distancing and other precautions deemed safe by its local health authority. 

As Clay Shupak relates in a wide‐​ranging piece at the Wake Forest Law Journal, the clause is also the subject of a pending challenge by a New Bern ophthalmologist to the state’s “certificate of need” law as applied. Such laws require those who seek to build or offer a new facility or service, in this case a specialized operating room in a small community, to prove against objections that there is (as somehow defined) an economic need for it. Carolina Journal covers remarks about the clause by North Carolina Supreme Court Justice Phil Berger, Jr. here and here.

Given North Carolina’s real‐​world experience with the clause, which some might see as inspiring and hardly anyone seems to regard as disastrous, it wouldn’t surprise me if there developed moves to add similar language by amendment to other states’ constitutions.

Patrick G. Eddington

Get ready for another one of those “Congress is trying to pull a fast one” stories … and if you work for or financially support one or more nonprofits, you’ll want to read on. Congressional sources have informed Cato that an extremely controversial tax code bill that could be used to target politically disfavored nonprofits that has already passed the House may get a Senate vote this week. First, some important background information.

The very same day (April 15) the House was voting on the atrocious Foreign Intelligence Surveillance Act (FISA) “reform” bill, it also passed the very innocuous sounding H.R. 6408, the purpose of which is “To amend the Internal Revenue Code of 1986 to terminate the tax‐​exempt status of terrorist‐​supporting organizations.”

The bill’s author, Rep. David Kustoff (R‑TN), issued a press release when his bill passed the House that claimed the following:

Recent reports indicate there are U.S.-based nonprofits that are suspected of providing support and funding to terrorist groups.

The gentleman from Tennessee offered exactly zero evidence to support the statement, a fact confirmed by the Charity and Security Network, itself a nonprofit.

Yet despite failing to produce any data from the FBI, DHS, or the Office of the Director of National Intelligence (ODNI) to back up his claim, Kustoff managed to convince his House colleagues to pass a bill that would allow the Secretary of the Treasury to designate a US nonprofit as a “terrorist supporting organization,” if it is found to have provided “material support or resources” or “training” or “expert advice or assistance” to a designated foreign terrorist organization per 18 U.S.C. § 2339A.

The actual target of this legislation is not ISIS or Al Qaeda, but nonprofit groups that are supportive of a cease‐​fire in the Israel‐​Hamas war or who are otherwise rhetorically supportive of the Palestinians or even seeking to provide humanitarian aid to people in Gaza.

As Matthew Petti at Rea​son​.com recently wrote, “Under the proposed bill, murky innuendo could be enough to target pro‐​Palestinian groups. But it likely wouldn’t stop there.” I agree, and there’s already a well‐​documented and thoroughly dishonorable history to this practice that should serve as a warning about the very dark place this current episode may take us.

Just shy of five years ago, I provided a little bit of a history lesson on previous episodes involving the political weaponization of the tax code or business or organizational records by Congress. In that instance, it was the infamous House Un‐​American Activities Committee (HUAC) strong‐​arming then‐​President Franklin D. Roosevelt into giving the committee access to tax and business records on individuals and organizations HUAC believed were Soviet controlled or influenced. As I noted then:

Between 1938 and 1975 (when it was finally abolished by Congress), HUAC either attempted to obtain or actually utilized sensitive personal information on American citizens to conduct anti‐​communist persecutions that destroyed the professional and personal lives of thousands of citizens.

This time it appears that Congress wants to outsource the tax status‐​centric political repression to the executive branch, specifically the Treasury Department. But whether it’s a presidential administration or Congress itself doing it, demanding the tax, business, or organizational records of a civil society organization that takes unpopular political positions to chill the speech or activities of that organization is pure political repression.

Indeed, Arab and Muslim Americans have lived under a state of de facto political siege since the 9/11 attacks, something I and many other scholars have written about at length. If Kustoff’s bill becomes law, it will not only likely intensify government repression against Palestinian Americans, it will also set a dangerous legislative and policy precedent that could be used against other nonprofits deemed a “terrorist supporting organization” even in the absence of concrete, verified evidence to that effect.

Is that the kind of thing we want the Treasury Department to do? Is that the kind of country we want to live in?

David J. Bier

The employer‐​sponsored green card system is so backlogged, delayed, and overregulated that it is nearly impossible for foreign workers to get through without first obtaining a work visa. The H‑1B visa is the main option available. All applicants must have at least a bachelor’s degree, and two‐​thirds have a master’s degree or higher. H‑1B workers have wages in the top 10 percent of workers nationwide. Unfortunately, the H‑1B visa also has a low cap of 85,000 for most types of jobs.

As a result of the cap, the government has awarded H‑1B visas through a lottery since 2014. This week, US Citizenship and Immigration Services (USCIS) announced that it received nearly half a million lottery entries for the 2025 fiscal year (which starts in October). It selected 120,603 individuals to move on to the application stage (25.6 percent), and of those selectees, just 85,000 will receive a visa.

Ultimately, just 18.1 percent of lottery entries will result in someone receiving H‑1B status. In other words, 82 percent of these skilled workers will be denied permission to work in the United States in H‑1B status.

For the FY 2021 lottery, the government introduced a new application method. Rather than submitting a complicated petition containing all the information necessary to obtain an H‑1B visa, it established a streamlined process. This allowed employers to register for the lottery for a small fee and provide only basic information about the worker. While this change was an improvement, it also made it easier for workers to line up multiple employers to submit registrations on their behalf, as there was little cost to doing so. This phenomenon reached its peak in FY 2024, when a majority of the 758,994 entrants had multiple employers petitioning for them, which increased their odds of winning.

For the FY 2025 lottery, the government changed the system again. This year’s lottery selected specific workers rather than employers, so having multiple submissions did not affect a worker’s odds of winning. Every worker received equal odds. Though it may have disadvantaged the go‐​getters who actively pursued multiple offers, this change was certainly fair. However, some people anticipated that it would improve the overall odds of winning the lottery, which it did not.

The H‑1B lottery has always proceeded in three stages. First, the applicant enters the lottery. Second, the applicant is selected. Third, the applicant’s employer must petition for them. To account for various factors like applicants declining the offer, employers withdrawing, or denials of workers or employers, the government always selects more lottery “winners” than the final number that receive a visa.

In the FY 2024 lottery, there were so many entries with multiple job offers that the government selected more than twice as many winners as was necessary to fill the visa cap. With fewer double entries in FY 2025, it selected 66,000 fewer winners, so in both years, the overall odds of selection were just 24.8 percent for 2024 and 25.6 percent for 2025. By the third step, the number will fall to just 18.1 percent getting a visa (Figure 2).

The odds certainly did increase for individuals without multiple registrations submitted on their behalf, but the odds aren’t in anyone’s favor. There is nothing special about the number 85,000. Given the average wage for H‑1B workers at $130,000 per year, Congress is randomly turning away at least $61 billion in economic contributions for no reason. I have previously written about the kinds of reforms that Congress should be urgently pursuing for Regulation magazine:

The H‑1B visa’s restrictions are harming U.S. competitiveness and innovation. Ideally, the H‑1B cap should simply be eliminated, but at a minimum it should be updated to reflect the current demand for visas, which is about four times the available visas. Moreover, visas that go unused or are revoked should be added back to the cap, and when an H‑1B worker receives a green card, the H‑1B slot should go to another applicant. These changes would put the focus on at least maintaining the stock of H‑1B workers rather than on the annual number of visas issued.

The longer Congress waits, the worse the situation becomes. Talent will increasingly flock to countries like Canada, Australia, the United Kingdom, Germany, India, China, and others that would welcome these workers. Where talent gravitates, much of the innovation, investment, and economic growth will follow suit.

David Inserra

Senator Ted Cruz’s (R‑TX) staff on the Senate Commerce Committee recently published a report that investigates several instances of what it calls “Online Service Providers … silencing conservatives” by relying “upon biased left‐​wing organizations.” While these specific examples may be viewed as mere anecdotes, the report dives deeper than many other accusations of political bias and tries to understand how exactly these companies reached their decisions to de‐​platform various conservatives. 

The report concludes that “The ideal solution is for market forces to correct Online Service Providers’ discriminatory policies.” While the report also embraces some limited legislative fixes that do not live up to this ideal, its emphasis on understanding how content decisions are made and how the market can help address real or perceived bias is constructive. In this same vein, I will be publishing a paper next month that explains how social media content moderation works and how the market and civil society can support greater expression and greater choice online.

It’s worth noting that it is and should be the right of such companies to set their own policies and deny their services to those with whom they disagree. But, the report argues, then we should dispense with the pretense that these are neutral companies trying to offer their services to most Americans so that the market can satisfy those no longer being served by these large tech companies. And unlike many other criticisms of tech company decisions, the report spoke directly with the companies to understand how and why these decisions were made.

The first highlighted decision was by Slack, a workplace and communications system, to suspend the right‐​wing social media influencer Libs of TikTok because of its various posts and reporting about what myriad LGBT organizations and activists were themselves posting online. The report notes that the deplatforming effectively caused Libs of TikTok employees to lose their prior communications on Slack, causing significant disruption to their organization’s operations.

The report then looks at Eventbrite’s cancellation of various events by groups like Young Americas Foundation, College Republicans, and others hosting conservative commentator Matt Walsh or his documentary What is a Woman? for questioning a progressive view of sex and gender. Similarly, Eventbrite cancelled an event featuring Riley Gaines, an elite college swimmer, for her views regarding transgender athletes participating in female sports. These cancellations often caused significant disruption and costs to the event organizers.

The report also notes other instances of deplatforming including the immigration‐​critical Foundation for American Immigration Reform (FAIR), the Independent Woman’s Forum, and other organizations that were excluded by these and other online service providers because of mainstream political and social views on important issues of our day.

The enforcement described in this report is different than your typical social media content moderation. When dealing with billions of pieces of content, AI‐​powered enforcement tools are necessary to weed through the sheer amount of material, often paired with large contingents of human moderators who must make relatively quick moderation decisions. This report describes more comprehensive reviews by expert and executive trust and safety teams that are looking at a broad range of factors.

This was one of my roles when I was a member of Meta’s content policy team and was often reserved for the most difficult and socially or politically sensitive content. And while using highly trained and expert moderation for difficult decisions makes sense, it also risks allowing external and internal biases to influence the outcome. 

For example, the committee report identified outside organizations that may exert significant influence over not just the writing of various policies but also how those policies are enforced. Eventbrite testified to the committee that it “relies on third‐​party sources, including Southern Poverty Law Center (SPLC) and the Anti‐​Defamation League, in determining whether an organization’s event violates the Eventbrite Community Guidelines.” Similarly, Slack told the committee that “it generally relies on “third‐​party experts” and “industry‐​recognized resources” in enforcing its policies. 

The way this works in practice is that external and generally progressive “experts” and “partners,” frequently, persistently, and aggressively lobby for certain users, pieces, or types of content to be removed as hateful or otherwise harmful. It is rare for right‐​leaning or libertarian groups to be as trusted internally or to report as much content as these progressive “partners.” 

Often, the content did not violate the clear letter of the policies, but pressure was applied by these organizations through media campaigns, organized boycotts, or one‐​sided research designed to make companies appear complicit in some harmful content. The result of such external pressure, together with receptive internal teams, is that sometimes companies cave or willingly support the suppression of certain viewpoints. They invoke high‐​level principles or certain contexts rather than specific policy lines or may even change the policy to align with activist demands.

This can be seen at work in this committee report, in which the service providers often avoid detailing how a user specifically violated their policies but instead refer to vague principles or contextual information they believe to be important. For example:

When asked if they were removing events featuring Riley Gaines because she made an X post about how women lack a Y chromosome, Eventbrite said the post “speaks for itself.” 

“[W]hen asked multiple times if Eventbrite would remove another event concerning women’s sports that featured Riley Gaines, Eventbrite dodged the question.”

Libs of TikTok was “problematic” on Slack because of its “specific audience.”

Eventbrite cited “the overall tone and message” of the trailer for the What is a Woman? documentary … “in combination with Matt Walsh’s related public statements,” rather than anything specifically violating in the film, for removing various organization’s viewing parties of the film or even unrelated events with Matt Walsh.

FAIR was de‐​platformed by Slack for being “affiliated with a known hate group” (likely referencing lists maintained by groups like the SPLC).

As I noted earlier, these organizations can be as vague, imperfect, or biased as they want in creating or enforcing their rules. And as the report concludes, “First Amendment protections generally do not apply to actions by private companies, who have the freedom to associate and do business with the customers they choose.” Therefore, “the ideal solution is for market forces to correct Online Service Providers’ discriminatory policies. If Online Service Providers continue to cancel conservative organizations, it should create a new demand for Online Service Providers that service the conservative market.”

This is exactly right. Nothing is stopping new service providers from entering into these fields to provide these important services—except for the threat of additional tech regulation by those on the right and the left. Those who want more expression online should make the case to current and prospective service providers that policies limiting significant amounts of socially important speech may not be serving these companies or the broader society well. Those who want more restricted speech have no problem aggressively making their case for less speech, leveraging increasing sympathy in academia, government, and society with their views as various metrics point to increasing conflict over the value of free expression. 

Those of us who want to see greater free expression must make the case not only for the First Amendment legal protection from government censorship but also make the case for a culture of free expression, including to the companies that are limiting vibrant discussions of important social issues. 

The research done by the Senate report is part of this effort to promote a broader culture of free expression, but it goes a bit too far in seeking legislation that makes some significant demands of private companies. Specifically, the report calls for intrusive transparency from companies regarding their policies and enforcement actions. While many of these recommendations may be best practices that I think companies would benefit from employing, I also know that sometimes companies don’t want to provide complete transparency in their rules. A common reason is to prevent adversarial actors from gaming their policies.

Similarly, requiring specific transparency around why users are being punished or deplatformed appeals to our desire for due process and fairness, but companies don’t need a reason, much less a clear or good reason, to remove users from their services. Ultimately, their lack of transparency and due process are creating bad experiences with their products that, unchecked, will likely result in alternative solutions emerging in the marketplace, a reality that the Senate report elsewhere understands.

Judge Learned Hand famously stated, “Liberty lies in the hearts of men and women; when it dies there, no constitution, no law, no court can even do much to help it. While it lies there it needs no constitution, no law, no court to save it.” We must remind our fellow Americans why a culture of free expression is so important and how the free market can address bias and discrimination.

Brian Chau

In general, the policy discussion around Artificial Intelligence (AI) underestimates the diversity and vulnerability of the AI startup ecosystem. The reality is often different than what policymakers think. Without understanding the startup ecosystem, the resulting policy could lead to unnecessary interventions into competitive markets that disrupt innovation in the free market.

So what might policymakers need to better understand about the current state of the AI startup ecosystem?

Policy analysis that focuses on the effects on industry leaders rather than on startups underestimates the damage done to AI development by harsher regulatory approaches, such as the Biden Executive Order on Artificial Intelligence. For example, the National Telecommunications and Information Administration (NTIA) is set to impose auditing and content regulation on “all stages of the value chain.” This means that instead of focusing narrowly on the areas where harms actually occur, the NTIA treats all AI developers as preemptively suspect.

Consequently, developers whose products have no risk of causing the harm the NTIA identifies are still forced to censor the output of their software, conduct time‐​consuming audits, and be delayed by inefficient regulatory approval processes.

This regulation would make the experimentation process far more time‐​consuming by increasing compliance costs, restricting what instructions AI models will respond to, and possibly harming the underlying efficacy of AI models. This results in more startups failing because they cannot acquire the capital to continue operating, there are longer timelines to complete research and development for the ones that persist, and there is a less effective end product. The NTIA’s proposed regulations would also have an anticompetitive effect, as they impose fixed costs that disproportionately affect smaller businesses with less time and capital to comply with them.

AI is often equated with its industry leaders, such as OpenAI, Anthropic, Google or Meta. However, just as assembling a car requires many suppliers and subcontractors, user‐​end AI products often consist of many specialized layers. Examples include producing the hardware required to run AI models, curating specialized data used as an input to AI models, computer intensive pre‐​training to produce a general language model such as GPT‑4, or fine‐​tuning a general model to improve performance on a specific task. The following chart is a map of these categories, which captures the complicated and competitive landscapes of each AI sub‐​industry.

Click here to see a larger image of the chart. 

Competition is increasing on each of these layers as AI becomes a more well‐​funded industry. As basic tools for every part of the AI pipeline become more widely available, startups can conduct thorough experiments on a narrower niche of the AI market.

Many startups will use a base model, such as OpenAI’s GPT‑4 or Meta’s Llama 3, as an input to their product. They experiment with a variety of base models, infrastructure, hardware, and optimization techniques to create the best product for their use case, leading to a complicated network of de‐​facto suppliers and subcontractors. Just as software companies now rely on other software for web development, databases, networking, or cloud services, AI companies will rely on increasingly specialized vendors for different components of AI models.

This process of market specialization is best demonstrated by the practical experience of startups. Proemial AI is a company that narrows down GPT‑4, sacrificing generality to make it the best at doing one specific thing: communicating scientific papers. “We were just surprised at how bad [unmodified] GPT4 was, because we were constraining it in interesting ways,” said founder Geet Khosla. Scientists often struggle to anticipate what questions an ordinary user would have about their work. By being able to respond to a user’s questions and vary its depth of explanation based on the experience of each individual user, Proemial would be able to make scientific work more accessible.

While most AI models require expensive hardware to run, Hal​tia​.AI is creating one that doesn’t. “We are taking the Mistral‐​7B model and putting a [version] on the iPhone,” said Hal​tia​.AI founder Arto Bendiken. This allows users to get an AI response much faster, at a lower cost, and even when they aren’t connected to the internet. This highlights the tradeoff between accuracy and convenience, which presents another opportunity for startups to compete.

Lastly, consider Superflows, a company that assembles a chain of Large Language Models (LLMs), ChatGPT, to explain a company’s software to a new programmer. “There’s a relatively long chain of prompts we use … get one LLM to describe endpoints … another LLM reads it,” said Superflows co‐​founder Matthew Phillips. His company experiments with AI models, their own modified models, and ways of formatting user input. “There’s a process of evaluating which LLM has the best ratio of quality to speed.”

These extensive experiments on base models and techniques carried out by startups require flexibility to operate. The NTIA’s proposed regulations impede them in several ways that are particularly impactful on smaller businesses. The most obvious issue is compliance costs for auditing or content restrictions, which each company would have to adapt their end product to. This serves as a barrier to entry for new companies, which may not have the resources to comply.

The result of content restrictions on their base‐​model suppliers would also make experimentation more difficult. Restrictions on model content can result in denying user requests that are not actually harmful. Overcoming these restrictions in non‐​harmful use cases sacrifices crucial engineering hours. Finally, there is some evidence that the techniques used to achieve content restrictions can reduce the overall performance or flexibility of AI models.

Brian Chau is the executive director of Alliance for the Future. He is a former machine learning engineer and International Olympiad in Informatics gold medalist.

Jeffrey A. Singer

The Drug Enforcement Administration announced today that it intends to reschedule cannabis as Schedule III on its schedule of controlled substances. For the past half‐​century, the DEA classified cannabis as Schedule I: “No currently accepted medical use and high potential for abuse.”

Of course, no serious person would argue that cannabis has “no currently accepted medical use.” As far back as 1916, Sir William Osler, the so‐​called “father of modern medicine,” recommended cannabis as the “drug of choice” for treating migraines. But cannabis’s history of “accepted medical use” dates back to at least 2800 B.C.

President Biden had asked the Department of Health and Human Services to review cannabis classification in 2022, and last summer HHS recommended that the DEA reclassify cannabis to Schedule III: “Drugs with a moderate to low potential for physical and psychological dependence.”

The good news is that the federal cops practicing medicine—the DEA—finally recognize that cannabis has medicinal uses. Rescheduling should make it easier for patients to obtain the drug with a prescription in the states that have not yet legalized medicinal cannabis. To date, 38 states and the District of Columbia have legalized medicinal cannabis. It will also make it easier for researchers to perform high‐​quality studies on the plant’s medicinal uses. And it will make it easier for cannabis retailers to take federal tax deductions from which the law has barred them.

The bad news is that it is still federally illegal for people to use cannabis recreationally. With rescheduling, the only way people will be federally permitted to purchase and consume cannabis will be if a health care practitioner prescribes it to them. This should be welcome news to recreational cannabis purveyors in the unregulated black market in the 26 states where recreational cannabis remains illegal.

The medical profession has long recognized that alcohol has medicinal uses. There is evidence that consuming moderate amounts of alcohol may be good for you. Alcohol can also be addictive and cause serious health problems, including cirrhosis, gastrointestinal cancers, cardiomyopathy, and encephalopathy. Yet, after their disastrous experience with alcohol prohibition, federal regulators have never sought to list alcohol as a controlled substance. People don’t need a doctor’s prescription to purchase a product from their local liquor store.

Tobacco has no currently known medicinal use but many known harmful effects. And, while relatively harmless, the nicotine in tobacco smoke can addict smokers and expose them to tobacco smoke’s harmful components. Yet the DEA has never listed cigars or cigarettes on its schedule of controlled substances, even though one can argue that they have “no currently accepted medical use and a high potential for abuse.”

I am not arguing for the DEA to add alcohol and tobacco to its list of controlled substances. On the contrary, I am calling for the DEA to remove a much less harmful—and more medicinally useful—product from that list.

Rescheduling cannabis is nice. It’s a small step in the right direction. But it doesn’t go far enough. The DEA should de‐​schedule cannabis.

Jennifer Huddleston

Last week, some headlines stated that the US had passed a “TikTok ban” as part of the foreign aid package, likely eliciting panic from many of the millions of American individuals and businesses who use the popular social media app. But the actual proposal is more nuanced, and the exact questions of what will happen are likely to take months—if not years—to settle. So, what should Americans expect now that the proposal is law?

In the next few months users are unlikely to notice changes.

Calling the proposal that was signed into law a “TikTok ban” is a bit of a misnomer although that could end up being a result. The actual proposal that was included in the foreign aid package requires certain apps designated by the US government as being controlled by a foreign adversary to divest in a way approved by the US government as removing that influence within a certain time frame. If they fail to do so in the time frame, the proposal prohibits the distribution, updating, or maintenance of such apps.

While the proposal could apply to other companies as discussed below, it explicitly names TikTok and parent company ByteDance as a company to which its terms apply. In short, the signing of the bill started a clock that forces TikTok to engage in a sale or face a ban. This means that the company would have roughly nine months (a rather short time frame) to find a buyer for some portion of the app that satisfies the government’s concerns or face a ban.

It remains unclear exactly what elements of the social media app would need to be sold to respond to these concerns or what, if any, buyer the government would consider as alleviating these challenges without facing additional regulatory scrutiny.

To be clear, as I have discussed in other work, there are many concerns about the impact this proposal will have on TikTok’s American users and their speech rights. Still, for the next few months, the legal status of the app does not really change, and most users are unlikely to experience changes related to the law. In general, users should still be able to update and use the app and see it available on app stores as they did before the law’s passage.

Litigation is almost certain.

TikTok has already said it will challenge the law in court; however, no case has yet been filed. This challenge is likely to raise several concerns, including First Amendment questions, takings‐​related issues, and possibly bill of attainder.

Particularly should a sale fail in the time frame, TikTok users or creators could challenge the law for the impact it would have on their speech and their business. I have explored the legal issues particularly related to the First Amendment rights of these users in previous work.

While a divest or ban proposal is distinct from prior attempts at a ban, precedents around bans indicate that the courts will likely be skeptical of such a significant action when less speech‐​restrictive means like disclosure, data audits, or data localization could respond to national security concerns. Past examples of legal challenges to TikTok bans, including the executive order during the Trump administration and the Montana state‐​level ban of 2023, have resulted in these bans being struck down or enjoined.

So, what will the court have to consider in its analysis of likely constitutional questions in a challenge to a ban? Much of the discourse focuses on the First Amendment‐​related, free expression concerns. This means courts will consider if the proposal merits intermediate or strict scrutiny and whether the government has made a sufficient argument on its national security grounds compared to the impact on speech.[1] Additionally, the courts will also likely consider if the proposal is content‐​neutral or a content‐​based restriction and/​or whether it constitutes a prior restraint.

The litigation could mean it is an even longer time before the implementation of any divest‐​or‐​ban timeline. The parties will likely request a preliminary injunction during litigation, which could at least pause any impending bans. A preliminary injunction would be likely if, for example, the court found there was a high likelihood of harm to speech while the case was pending.

Bigger than TikTok.

While this proposal names TikTok and has some limiting principles, it could be applied to other apps as well. As many have noted, popular e‑commerce services like Shein and Temu, which also have ties to China, could likely be covered as well.

But the proposal could extend much farther and be used to target specific companies an administration might disfavor. For example, Parler at one point relied on a Russian DDoS (distributed denial of service) guard service, and conservative video platform Rumble hosts Russian state media content. Could this be used by an administration to label those sites as controlled by a foreign adversary? As Senator Rand Paul (R‑KY) discussed in a Reason op‐​ed, without clear limits, this could allow the government to force sales or changes from many more companies than one might initially think with little due process to respond to such a classification.

Conclusion

The average TikTok user may not experience changes immediately as a result of the signing of the often mischaracterized “TikTok ban,” but the actual impact of the law is more significant than TikTok. Whether one is a TikTok user or not, this new law could open the door to significant executive intervention in the technology market and it raises First Amendment concerns that merit attention.

[1] A federal court issued a preliminary injunction to the Montana TikTok ban stating it was likely to fail even intermediate scrutiny but did not reach a conclusion on the appropriate level of scrutiny.

David J. Bier

To sponsor a foreign worker for a green card (i.e., permanent resident status in the United States), an employer must first receive an approved permanent labor certification from the Department of Labor (DOL). This certification states that the green card will not adversely affect US workers. The statute requiring the labor certification provides no details about what this process should look like and, into this void, the DOL has unleashed an avalanche of complicated requirements, forcing employers to recruit US workers before they can receive a certification.

But the DOL will waive many of these complicated requirements in the case of a shortage occupation. The DOL’s shortage occupation list is known as Schedule A. It was last updated in 1991, and it contains just two “shortage” occupations: physical therapists and registered nurses (as well as people with “exceptional ability,” but this classification is reportedly almost never used, though this might change thanks to a new US Citizenship and Immigration Services’ policy change announced this month). For the last 33 years, DOL has never recognized any other occupations as being in short supply. But now DOL is requesting that the public weigh in on whether to expand the list.

The permanent labor certification should not exist. Green card holders do not harm US workers because foreign workers consume goods and services, which increases demand for US workers elsewhere in the economy. Green card holders can’t be paid below market wages because they can simply leave to go to a job offering the market wage. This bureaucratic protectionism ultimately hurts US workers to the extent that it keeps out skilled workers who will increase demand for their services in other jobs elsewhere in the economy.

The permanent labor certification as it exists is an unnecessary bureaucratic nightmare. In 2004, when the government announced it would be streamlining the labor certification, it stated that most applications would receive a decision between 45 and 60 days under what it calls “non‐​audit review,” which is supposed to be a streamlined approval process.

Instead of 60 days, these applications are now taking an astounding 397 days—up from 96 days in the second quarter of 2019. Instead of meeting the 60‐​day limit announced in the regulation, the wait time is growing by almost 60 days every six months.

Naturally, the result is that a huge backlog has developed, increasing from about 27,000 in 2019 to over 161,000 in the first quarter of FY 2024. This is now more than an entire year’s worth of filings and adjudications. It’s likely that at the current pace, none of the filings being submitted right now will be reviewed by the DOL for a year and a half or more.

The DOL is not increasing adjudications in response to this crisis (Figure 2). From calendar year 2019 to 2023, DOL has not adjudicated as many applications as in 2018. In total, just keeping up with the 2018 pace would have reduced a cumulative 61,000 backlogged cases. That pace would have been insufficient to keep up with demand, but it would be better than where the agency is now.

Given that it has proven incapable of increasing adjudications through the individualized procedure, DOL must update Schedule A to meet its legal obligations to issue certifications and follow its own regulations.

It’s also important to locate the permanent labor certification in the full context of the employer‐​sponsored immigration process. The first step of the labor certification—which even Schedule A applicants must perform—is to receive a prevailing wage determination from the DOL. This is also unnecessary because DOL publishes the prevailing wages for each occupation on its website. Naturally, DOL is doing historically bad at processing these applications as well, with the average application taking over six months.

Then, the average employer takes another six months from the prevailing wage determination to recruit and prepare to file a labor certification application. Altogether, DOL’s procedures are taking an astounding 800 days for the average employer in 2024 (Figure 3). That’s not including lawyer consultations before the process begins. Schedule A cuts over 600 days from this process.

Although DOL’s processing times have escalated in recent years, the permanent labor certification has been a convoluted disaster since the current iteration began in 1977. Until 1965, the permanent labor certification only existed as a veto in cases where the government could show that workers would be harmed, such as during a strike. After 1965, the language shifted the burden to the immigrants to prove that they would not harm US workers. However, in 1967, the DOL clarified that it would not be using the new system similarly to the old system.

As Figure 4 shows, the type of labor certification is a key predictor of whether employer‐​sponsored immigrants must obtain another temporary visa first before the employer tries to run through this process. This two‐​step process has become a necessity for employers because the labor certification is time‐​consuming and complicated. Employers cannot sit around for years waiting for the DOL to approve them, so the DOL has funneled all these permanent employees into the temporary system unnecessarily. This reduces visas available in the temporary system while limiting the rights of the workers who come.

Despite this monumental bureaucratic morass, DOL denies few permanent labor certifications. From October 2014 to December 2023, DOL denied a total of just 50,427 permanent labor certification applications compared to 892,380 approvals. DOL is reviewing 19 applications to get a single denial. This fails any reasonable cost‐​benefit test. Employers are spending tens of thousands of dollars to go through this process. The permanent labor certification needs fundamental reforms to streamline it. At a minimum, DOL should conduct an in‐​depth audit of a random percentage of employers and simply approve the rest, which is effectively the system that the 2004 regulation promised.

However, since DOL cannot figure out how to efficiently handle them, it should put more occupations in Schedule A. Table 1 lists 83 occupations accounting for two‐​thirds of all labor certifications, which have denial rates of 2 percent or less (including only occupational codes with at least 20 applications). DOL is reviewing 57 applications for each denial in these occupations. This is unjustifiable. Moving these 83 occupations to schedule A would speed up the other one‐​third of applications and give the agency more time to review the applications with higher denial rates. These occupations have demonstrated that there is a shortage, so there is no need to subject them to an individualized review.

This is not the only way that DOL should consider whether an occupation is a shortage occupation, but it is a useful place to start. This is something that DOL has directly measured over a long period and bears directly on whether it is worth DOL’s limited resources to review these particular applications. Of course, the employment situation in the United States is not static, but the unemployment rate has almost no relevance to these highly specialized jobs. In 2020, the unemployment rate increased without any change in the labor certification approval rate.

To avoid a perpetual Schedule A list, however, the DOL could reinstate individualized review to see if the situation has changed if the yearly average unemployment rate in the occupation has increased by more than two percentage points since it was last listed as Schedule A. The DOL could then reassess these occupations again after six months to see if 98 percent or more are still being approved.

DOL’s goal should always be to optimize its resources and avoid the lengthy delays that harm US companies that file labor certification applications. According to DOL data, these companies employ over 84 million people, and harming these companies only harms their existing employees who work alongside immigrant workers. It also hurts countless US consumers who use the products and services that these immigrants produce in the United States. DOL must immediately address its permanent labor certification processing and expand Schedule A.