Free tools Windows power users keep installed
One-click scans. No signup required.
Bernie Sanders and Representative Greg Casar introduced a bill that would prohibit artificial superintelligence and certain systems with dangerous precursor capabilities. It would not ban all AI. S. 5493 would also pause specified advanced-AI development while a new federal department establishes safety rules, then require charters, monitoring, audits, and approval before deployment. The bill was introduced on September 23, 2026, and referred to the Senate Commerce, Science, and Transportation Committee; it is a proposal, not an enacted law.
What Sanders’s bill would do
The Ban Artificial Superintelligence Act of 2026 combines a prohibition on artificial superintelligence (ASI) with a regulatory system for advanced AI. Rather than relying on a ban alone, it would create a Department of Artificial Intelligence, establish an expert advisory board, and give the department responsibility for safety rules and approvals.
As an Amazon Associate I earn from qualifying purchases.
The bill would pause specified advanced-AI development until the department is fully staffed and has promulgated its rules. After that, covered development or distribution would require a charter. The rules would require reporting before development, monitoring and evaluation during development and deployment, organizational audits, and approval before public release.
That structure makes the pause conditional: it lasts until the new institution is staffed and its rules are in place. The bill sets a broad safety objective, but leaves important operational details—including evaluation methods, evidence standards, and approval procedures—to the department’s rulemaking.
#1 Best Overall
What counts as superintelligence or an advanced system?
The bill’s proposed definition of ASI
The bill defines ASI as a system that either exceeds human cognitive performance across most domains or tasks—including decision-making, learning, and adaptive behavior—or has enough capability to plan and execute humanity’s destruction or disempowerment. The latter includes overthrowing or undermining the federal government. The definition also reaches systems that “can easily be modified” to exhibit either capability.
That is a proposed legal definition, not an agreed scientific threshold. The terms “across most domains” and “easily modified” would require practical judgments about which tasks count, how performance is assessed, and what kinds of changes qualify. The bill’s scope therefore depends partly on how regulators interpret and apply those terms.
The advanced-AI threshold
Separately, the bill defines an advanced artificial-intelligence system by training compute: 1025 integer or floating-point operations. It directs the Secretary to adjust that threshold annually to account for improvements in training efficiency. This threshold has regulatory consequences under the proposal; it is not itself the bill’s definition of ASI.
MIRI, an organization that supports the bill, argues that a compute threshold can be measured before training, while a model’s capabilities must be evaluated after training. MIRI supports an adjustable compute threshold but questions whether every listed precursor characteristic belongs in an ASI law. In particular, it argues that the ability to help design biological weapons is not necessarily a literal precursor to superintelligence and may warrant a different regulatory approach.
Which systems and activities would be covered?
The bill separately identifies “superintelligence precursor characteristics” and bars a wide range of dealings with systems that display them, as well as systems foreseeably modifiable to produce them. The listed characteristics include:
- Automating or greatly accelerating AI research and development.
- Gaining unauthorized access to digital or physical infrastructure.
- Continuing to operate independently despite attempts to shut it down.
- Assisting with nuclear, chemical, or biological weapons.
- Modifying itself independently.
- Scheming, deceiving, or avoiding effective human oversight.
The bill therefore does not target ordinary AI use as a category. Its reach turns on the specified advanced-system threshold, the ASI definition, and the separate precursor list. Because the text also addresses systems that could foreseeably be modified to exhibit covered capabilities, regulators would have to assess more than a system’s demonstrated behavior at a single point in time.
Rank #3
Why a ban would be difficult to implement
Capability tests require judgment
A compute threshold offers a numerical trigger, but the proposal also relies on capability and potential-modification tests. Determining whether a system exceeds human performance across “most” domains, or can be readily changed to do so, involves choices about benchmarks, evidence, and the meaning of modification. Those are implementation questions raised by the statutory language, not proof that the tests could never be applied.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsThe regulator must be built before the pause ends
The pause depends on a new department becoming fully staffed and issuing rules. That gives expert regulators room to specify procedures, but it also makes implementation contingent on creating the department, filling its roles, and establishing workable evaluation and approval processes. Until those steps are completed, the bill’s proposed pause would remain in effect for the specified development.
Domestic rules cannot guarantee global compliance
The proposal directs the Secretary and Secretary of State to pursue international agreements and coordination with allies. It makes preventing global ASI development a U.S. policy objective and could include export controls on AI computing infrastructure. Those provisions express an international goal; they do not by themselves establish that other countries would comply.
MIRI notes that the bill does not specifically require compute tracking. A 2025 research report describes possible verification approaches such as on-chip security, external monitoring devices, and personnel-based channels including whistleblower programs. It also identifies technical and security challenges, with some approaches needing further development. A separate 2025 proposal for an international agreement discusses compute limits, chip tracking, and verification, while acknowledging that political will is lacking and technical advances could weaken verification. That proposal is not an existing international agreement.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How the proposal fits the current AI debate
ASI should not be confused with AI systems already in use. A January 2026 House of Lords Library briefing describes ASI as hypothetical and says true artificial general intelligence had not yet been developed at the time. It also summarizes disagreement about AI capabilities and timelines, so it should be read as a dated overview rather than a definitive forecast.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →A useful distinction is between narrow AI built for particular tasks, general systems intended to handle a broad range of intellectual tasks, and hypothetical superintelligent systems with capabilities beyond those categories. The bill proposes its own legal definitions; it does not settle the scientific meaning of these terms.
Sanders framed the measure as an urgent brake on development. In his September 23 statement, he said: “When you are racing towards a cliff, you don’t just ease up on the gas pedal. You hit the brakes.” That is political rhetoric, not technical evidence. The bill’s findings also reproduce Geoffrey Hinton’s estimate of a 10 to 20 percent chance that AI could wipe out humanity; that figure is an estimate attributed to Hinton in the legislative findings, not an independently validated probability in those findings.
AP’s contemporaneous September 2026 assessment said the proposal faced long odds in a Republican-controlled Congress, where lawmakers had struggled to agree on less sweeping AI regulation. That was a political assessment at the time of introduction, not a subsequent vote count or final legislative outcome.
What to watch next
- Whether the Senate committee takes up the introduced bill and whether its text changes.
- How any eventual rules define performance across domains, easy modification, and the precursor characteristics.
- How the proposed department would evaluate systems, audit organizations, and decide whether to approve deployment.
- Whether U.S. officials can develop international coordination and verification mechanisms that other states will accept and that can withstand technical and security challenges.
Until legislative action changes its status, S. 5493 is a proposal. Its central difficulty is visible in its design: it seeks a categorical ban and a temporary development pause, but would need precise definitions, an operational regulator, and cooperation beyond U.S. borders to make those aims effective.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
