Sanders and Casar Introduce a Bill to Ban Superintelligence, With 20-Year Prison Terms

While the AI industry spent last week shipping frontier models and debating AGI thresholds, two lawmakers introduced the most aggressive AI bill in Congress. On September 3, Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act, which would permanently prohibit the development and deployment of AI systems matching or exceeding human cognitive performance, temporarily pause advanced AI development, and impose criminal penalties on violators of up to 20 years in prison, per Gary Marcus’s analysis of the text. It has no realistic path to passage in this Congress, but dismissing it misses what it signals.

The Ban Push Is No Longer a Fringe Position

The bill lands amid a visible political wave rather than in isolation. TIME reported on September 8 that lawmakers in both the US and UK are now openly discussing superintelligence bans, driven by fears of powerful systems escaping human control. And the release-timing connection to this week’s news is explicit: Sanders’s office rolled out the bill days before Anthropic researcher Jacob Coxon resigned warning that labs are “gambling with our lives” racing toward self-improving systems, followed by Anthropic’s own alignment science lead publicly endorsing a greater-than-10% decade estimate for human extinction from AI. The bill’s premise, that the risk justifies extreme intervention, is no longer argued only by people outside the labs.

Why It Matters Even If It Never Passes

Bills like this function as positioning for the version that can pass. The concrete policy landscape this bill competes with is the one this blog tracked all month: the G20’s non-binding Carolina Principles endorsed unanimously including China, and the EU’s hard-obligation model that just drew ChatGPT into VLOSE regulation. A criminal-penalty ban enters that spectrum as the maximalist anchor, which shifts what “moderate” means: against a 20-year prison term, licensing regimes, mandatory audits, and slowdown frameworks look like the reasonable middle. Frontier labs will lobby hard against exactly this framing, but every hearing it generates moves the US closer to the enforceable-oversight position that OpenAI’s own chief scientist requested. Watch the hearings, not the vote count.