By Aaron Scher; endorsed by Bourgon, Soares, and Yudkowsky on behalf of MIRI.
MIRI has been warning about the extinction threat from superintelligent AI for over two decades. Only recently has this danger become known in the policy world, and the proposed policies for dealing with the threat have to date been piecemeal and insufficient.
The Ban Artificial Superintelligence Act of 2026 is the first piece of legislation we’ve seen that stands a chance at stopping this threat. The Act is excellent but not perfect, and we discuss both what it gets right and what we'd tweak. We hereby endorse the Ban Artificial Superintelligence Act of 2026 because it directly confronts the extinction threat that humanity is facing and would codify the primary policy goal we think the world needs: a ban on the development of superintelligence.
What we like about the Act
Banning artificial superintelligence (ASI), or variants of such a plan, is the only effective solution to avoid the ASI threat, at least in the near term. Most other legislative proposals do not confront this threat head-on and thus would not be effective, even if implemented. For more on why we believe this, see Chapter 13 of If Anyone Builds It, Everyone Dies.
Banning ASI requires intervening before ASI is created at all. To this end, the bill requires that AI training pause if a model develops precursor characteristics, such as the capacity to automate AI R&D or the capacity to avoid human shutdown attempts. This approach is good because it avoids dancing too close to danger and it intervenes on one of the likely pathways to ASI: AIs automating AI development.
This bill includes a training compute threshold above which AI development requires a charter. Training compute thresholds are often a good tool for AI governance because they are measurable ahead of time (unlike evaluating the capabilities of a trained AI model). A major concern with training compute thresholds is that continued progress in AI algorithms and data means the same capability can be reached using less and less compute over time. The bill handles this issue by enabling the threshold to be adjusted to account for these more efficient methods, which is a good choice. It would be an improvement to allow these thresholds to be adjusted more frequently than annually, given that algorithmic progress can be very rapid.
We are glad to see the bill recognize and call for international solutions to the ASI threat. A domestic approach alone is insufficient. MIRI’s Technical Governance Team proposed an international agreement to prevent the premature creation of superintelligence, which is our current best guess at what such international coordination should look like. In line with the Act, we also strongly believe it should be “the policy of the United States to prevent the development of artificial superintelligence globally.”
The bill requires that advanced AI development pause until the relevant government agency is in place and staffed. This is a reasonable stop-gap measure because nobody knows for sure that continued AI development at the current frontier will not soon constitute a point of no return. We wish other policy proposals for AI risk would also take this conservative approach of “stop escalating AI capabilities until we have a better handle on the situation and an effective governance regime.”
Many MIRI staff want Earth's civilization to build ASI eventually, after the point where ASI can be created in a fashion known in advance to be non-ruinous. This is far beyond what present science can do by present methods, and the near term requires a complete, flat ASI ban.
Ways in which the Act falls short of perfection
The bill lacks tracking and monitoring of AI chips. We think such measures are important for preventing ASI from being developed anywhere on Earth, which is why they are a key focus of MIRI TGT’s international agreement. The bill is consistent with implementing policies in this direction (Sec. 15) but does not specifically mandate them. In a world with enough political will for this bill to pass, we expect Congress to pass other laws or amend this one to ensure sufficient compute tracking/monitoring.
In the long run, preventing the creation of superintelligence may involve restrictions on certain research (rather than just on the development of advanced AIs). Such restrictions are not present in the bill, which we think is a reasonable choice, but a stronger and more robust effort should likely involve such restrictions.
Some precursor characteristics in the bill seem directly related to ASI, such as the ability to automate AI R&D, while others do not seem like literal ASI "precursors", such as the ability to uplift biological weapon design. This second set of dangers likely merits a different optimal set of rules. For example, if sufficient safety, security, and access controls were in place and developers were licensed, we could imagine it being a good tradeoff for DeepMind to create non-ASI, specialized medical AIs, even if unauthorized access could uplift biological weapons. MIRI TGT’s international agreement was written to be configurable for exceptions like this if medical AI proves to be a key point of negotiation. Getting these exceptions right will be tricky, and it may be the correct decision for this bill to avoid trying to solve that problem. In the long run, however, we would consider such exceptions once an ASI ban is in effect. (If they cannot be carefully configured, they are not worth weakening a strong ASI ban and risking humanity.)
Overall, we think it would be an extraordinary improvement on present conditions if this bill became law. It is a substantial improvement over previously proposed legislation for handling ASI risk. Humanity needs to back off from building superintelligence until we have a better grasp on the alignment problem, and this bill would be a big step in that direction both in terms of direct impact and in the extent to which it takes serious issues seriously.
By Aaron Scher; endorsed by Bourgon, Soares, and Yudkowsky on behalf of MIRI.
MIRI has been warning about the extinction threat from superintelligent AI for over two decades. Only recently has this danger become known in the policy world, and the proposed policies for dealing with the threat have to date been piecemeal and insufficient.
The Ban Artificial Superintelligence Act of 2026 is the first piece of legislation we’ve seen that stands a chance at stopping this threat. The Act is excellent but not perfect, and we discuss both what it gets right and what we'd tweak. We hereby endorse the Ban Artificial Superintelligence Act of 2026 because it directly confronts the extinction threat that humanity is facing and would codify the primary policy goal we think the world needs: a ban on the development of superintelligence.
What we like about the Act
Ways in which the Act falls short of perfection
Overall, we think it would be an extraordinary improvement on present conditions if this bill became law. It is a substantial improvement over previously proposed legislation for handling ASI risk. Humanity needs to back off from building superintelligence until we have a better grasp on the alignment problem, and this bill would be a big step in that direction both in terms of direct impact and in the extent to which it takes serious issues seriously.