The U.S. AI Safety Institute is one of the few federal offices focused on studying risks in AI systems. Its work sits at the center of a larger question now facing Congress: should the office become a lasting part of the federal government, or remain tied to an executive order that a future president could undo?
The answer matters because the AISI is already connected to AI research, testing, and evaluation work with major AI companies and international partners. But its legal footing remains uncertain.
Why the AISI is vulnerable
The U.S. AI Safety Institute, known as AISI, was created in November 2023 as part of President Joe Biden’s AI Executive Order. It operates inside NIST, the Commerce Department agency that develops guidance for the deployment of different categories of technologies.
The office has several markers of a functioning federal initiative. It has a budget, a director, and a research partnership with the U.K. AI Safety Institute. Even so, its survival is not guaranteed, because it was established through an executive order rather than by Congress.
That distinction is the source of the current concern. If the AI Executive Order were repealed, the AISI could be wound down. Chris MacKenzie, senior director of communications at Americans for Responsible Innovation, described the risk directly to TechCrunch: “If another president were to come into office and repeal the AI Executive Order, they would dismantle the AISI,” adding, “And [Donald] Trump has promised to repeal the AI Executive Order. So Congress formally authorizing the AI Safety Institute would ensure its continued existence regardless of who’s in the White House.”
In practical terms, congressional authorization would shift the AISI from a single administration’s project toward a body with broader legal support. That would not automatically make it a powerful regulator, but it would make its existence less dependent on who controls the White House.
Funding is part of the fight
The AISI currently has a budget of around $10 million. The source article notes that this is relatively small when weighed against the concentration of major AI labs in Silicon Valley.
Supporters of authorization argue that Congress could give the office a more stable financial foundation. MacKenzie said appropriators in Congress tend to give higher budgeting priority to bodies that Congress has formally authorized, because those entities are understood to have broad backing and a longer-term role.
That funding question is closely tied to what the AISI is expected to do. Its work includes AI research, testing, and evaluation, and those activities become more important as companies build increasingly significant AI systems. A temporary or fragile office may struggle to plan for long-running technical work, while a formally authorized one could be better positioned to support continued benchmarks and coordination.
Industry and universities are pressing Congress
A coalition of over 60 companies, nonprofits, and universities has asked Congress to enact legislation codifying the AISI before the end of the year. The group includes OpenAI and Anthropic, both of which have signed agreements with the AISI to collaborate on AI research, testing, and evaluation.
Other major technology companies also signed the letter mentioned in the source article, including Microsoft, Google, Amazon, and IBM. Their support reflects a common view among think tanks, industry coalitions, and major AI firms: even if the AISI is not an enforcement-heavy body, it may be an important route toward AI benchmarks that can shape future policy.
The Senate and House have each advanced bipartisan bills to authorize AISI activities. Those bills have not moved without friction. Some conservative lawmakers have opposed parts of the proposals, including Sen. Ted Cruz (R-Texas), who has called for the Senate version of the AISI bill to pull back on diversity programs.
The debate therefore combines several issues at once:
- Whether AI safety work should be anchored in law rather than an executive order.
- Whether the AISI should receive more stable long-term funding.
- How Congress should define the scope of the office’s activities.
- Whether voluntary AI standards can still influence future AI policy.
What the AISI can and cannot do
The AISI is not described as a strong enforcement organization. Its standards are voluntary. That limits how directly it can compel companies to act.
But voluntary standards can still matter when they become reference points for companies, policymakers, and international partners. The source article identifies the AISI as a promising path toward AI benchmarks. Those benchmarks could help establish common ways to evaluate AI systems, even before any future policy is built around them.
This is why the AISI’s uncertain status is drawing attention. The office may not have hard regulatory authority, but it sits in a position where technical work, public policy, and industry cooperation overlap. Losing it could interrupt that work just as companies and governments are trying to define how AI systems should be assessed.
The global stakes
The U.S. is not acting alone on AI safety. At an AI summit in Seoul in May 2024, international leaders agreed to create a network of AI Safety Institutes. The network includes agencies from Japan, France, Germany, Italy, Singapore, South Korea, Australia, Canada, and the European Union, along with the U.K. and U.S.
That international context adds another layer to the congressional decision. Some interest groups worry that letting the AISI disappear would weaken U.S. leadership while other governments continue building their own AI safety capacity.
Jason Oxman, president and CEO of the Information Technology Industry Council, framed the issue in those terms in a statement: “As other governments quickly move ahead, members of Congress can ensure that the U.S. does not get left behind in the global AI race by permanently authorizing the AI Safety Institute and providing certainty for its critical role in advancing U.S. AI innovation and adoption.” He also urged Congress to act on the call from industry, civil society, and academia before the end of the year.
For now, the AISI remains active but exposed. Congress has bipartisan proposals on the table, a broad coalition asking for action, and a deadline shaped by the political calendar. The central question is no longer whether the U.S. has an AI Safety Institute. It is whether that institute will become durable enough to survive the next shift in power.