The Sanders-Casar ban bill triggers on capability to overpower the federal government
Bernie Sanders and Greg Casar want a federal ban on developing AI capable of overpowering the federal government. ControlAI, which says it advised their offices, argues the bill's mandatory pause on advanced AI is unnecessary.
The Engineer · Build desk
What happened
- Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act, which ControlAI calls the first American bill proposing to ban the development of superintelligent AI.
- ControlAI, which says it consulted the sponsors' offices, disagrees with the bill on its proposed mandatory pause on advanced AI development and on its treatment of precursors.
- ControlAI's own U.K. bill, which uses the same capability definition, was introduced in Parliament by Alex Sobel MP.
Why it matters
- What it limits constraint A trigger written as a capability gives a compliance team nothing to instrument. Whether a training program is lawful turns on who assesses the system's power over the federal government, and at what stage.
- What doesn't add up contradiction The group that says it helped shape the framework is arguing against the clause with the widest reach over ordinary model training, so the bill's closest ally and its sponsors are not aligned on scope.
- Who's exposed exposure A pause on advanced AI development reaches developers who would never claim to be building superintelligence, and the bill's own definition describes a different class of system entirely.
- What it sets up precedent With a U.S. bill and a U.K. bill built on the same capability test, that definition is now available for other legislatures to copy without resolving any technical threshold.
Next story · Invest · Oct 9
OpenAI and Anthropic executives war-game the backlash from an AI cyberattack insiders expect within a yearConfirmed4 publishers
A capability trigger and a compute threshold are two different compliance objects. ControlAI, the advocacy group that says it consulted with the sponsors' offices on the legislative framework [11], describes the bill's definition as turning on an AI system's capability to overpower the federal government [8]. The group's stated reason for liking that construction is that it targets development posing an extinction risk without settling in advance the technical details of how such an AI would work [9]. It also, ControlAI wrote, "keeps measures focused on the types of AI development that threaten the federal government, even if AI companies change the label they use to describe those projects" [10].
A definition like that leaves an engineering organisation without a counter to read. A FLOP budget emits a number you can log and cap. A capability to overpower a government is an assessment, and the statute's practical scope depends on who makes it and at what point in a training program.
The clause that would reach ordinary frontier work is a separate one. ControlAI's two disagreements with the bill are its proposed mandatory pause on advanced AI development and its treatment of precursors to superintelligent AI [3]. "We believe that the proposal's goals can be achieved with a more targeted approach, like the one in our draft bill," the group wrote [4]. Two parameters decide whether that pause touches a given run: what triggers it, and how long it holds. ControlAI's post does not include the bill's text or a bill number [5].
The bill also requires the United States to pursue international agreements to ban superintelligent AI development [2]. ControlAI, which supports the provision, discounts its force: it "is unlikely to bind the executive branch, since Presidents have long treated congressional directions on the conduct of foreign policy as advisory, and the question has never been settled in court," the group wrote [14].
This is still a bill, and the post reports nothing about committee referral, cosponsors or floor time. It does document a distribution channel. ControlAI says it has briefed more than 400 lawmakers across the United States, the United Kingdom, Canada and Germany in the last two years [6], which works out to more than 16 briefings a month on average [13]. Its UK bill was introduced in Parliament by Alex Sobel MP [7]. The group says it uses the same capability-to-overpower definition in both its U.S. draft bill and its U.K. bill [12].
What to watch
- The bill's operative text: what capability or scale triggers the mandatory pause on advanced AI development, and for how long.
- Whether ControlAI's narrower U.S. discussion draft is introduced as a competing vehicle rather than left as a comment on this one.
- Whether any cosponsors join an independent senator and a House Democrat on a prohibition bill.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+20
- Incentives70
- Confidence55
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Senator Bernie Sanders (I-Vt.) and Representative Greg Casar (D-Texas) introduced the Ban Artificial Superintelligence Act in Congress; ControlAI describes it as the first American bill to propose banning the development of superintelligent AI.
- [2]
The bill requires the United States to pursue international agreements to ban the development of superintelligent AI.
- [3]
ControlAI says its approach differs from the bill on the proposed mandatory pause on advanced AI development and on the treatment of precursors to superintelligent AI.
- [4]
ControlAI wrote: "we believe that the proposal's goals can be achieved with a more targeted approach, like the one in our draft bill."
- [5]
ControlAI's post does not include the bill's text, a bill number, or any report of committee referral, cosponsors or floor time.
- [6]
ControlAI says it has briefed over 400 lawmakers across the U.S., U.K., Canada and Germany in the last two years.
- [7]
ControlAI's U.K. bill was introduced in the U.K. Parliament by Alex Sobel MP, which ControlAI describes as the first ever bill introduced in the world to ban superintelligent AI.
- [8]
According to ControlAI, the bill defines superintelligent AI by its capability to overpower the federal government.
- [9]
ControlAI says the capability-based definition targets exactly the kind of AI development that poses a risk of extinction, without having to settle in advance the technical details of how such an AI would work.
- [10]
ControlAI wrote that the definition "keeps measures focused on the types of AI development that threaten the federal government, even if AI companies change the label they use to describe those projects."
- [11]
ControlAI says it consulted with the sponsors' offices on the Ban Artificial Superintelligence Act legislative framework.
- [12]
ControlAI says it uses the capability to overpower the federal government in the definition in both its U.S. draft bill and its U.K. bill.
- [13]
More than 400 lawmaker briefings over two years averages more than 16 a month.
- [14]
ControlAI wrote that the international provision "is unlikely to bind the executive branch, since Presidents have long treated congressional directions on the conduct of foreign policy as advisory, and the question has never been settled in court."
Sources
2 independent publishers whose own reporting we read for this story.
- lesswrong.comThe First American Bill to Ban Superintelligent AI Is Here
1 article · September 23, 2026
- the-decoder.comU.S. bill proposes permanent ban on artificial superintelligence and creation of new federal AI agency
1 article · September 24, 2026
Topics and entities
Topics
- Compute Thresholds
- Artificial Superintelligence