Sanders and Casar Want a Federal Agency to Ban Superintelligence Outright

Sanders and Casar Want a Federal Agency to Ban Superintelligence Outright

While chip controls leak, Washington is debating whether to ban the entire category of AI they're meant to stop. Sen. Bernie Sanders and Rep. Greg Casar introduced the Ban Artificial Superintelligence Act of 2026 in September, and it's still working through committee — a bill that would create a cabinet-level Department of AI with licensing authority over any model trained above 10^25 operations [4][5].

The bill doesn't mess around: a permanent ban on "superintelligence," defined as AI exceeding human cognition or capable of disempowering humanity, asset seizure for violators, up to 20 years in prison, and a full pause on advanced AI development until safety rules exist [5][6]. Sanders framed it in stark terms: "When you are racing towards a cliff, you don't just ease up on the gas pedal. You hit the brakes."

Whatever your view on the substance, the politics are notable — this is a serious legislative vehicle, not a fringe gesture, landing the same week evidence surfaces that China is already circumventing compute restrictions. The contrast is almost too on-the-nose: one side of the world debating whether to criminalize ambition, the other quietly buying the chips to pursue it anyway.

Fish Audio Ships No-Code Voice Agents With a 32-Second Deploy Time

Buried under the geopolitics is a product story that matters more for most of our readers day-to-day. Fish Audio launched Fish Agent on November 12 — a no-code platform for building AI voice agents that make and take calls with full-duplex, natural conversation, powered by their Fish Live 3 model [7]. Median time to deploy a working agent: 32 seconds.

It supports custom LLMs, tool calling, and webhooks, meaning it's not a toy — it's infrastructure meant to sit inside real business workflows, not demo reels. Fish Audio is giving away free credits at launch to drive adoption, and engagement has been strong, which tracks with everything we're seeing in voice AI this year: the bottleneck was never model quality, it was deployment friction.

For a space we build in directly, this is the trend to watch — not bigger models, but faster time-to-production. The gap between "impressive demo" and "agent in production" is closing to under a minute, and that changes what counts as a viable voice AI vendor.

EU Adopts Dublin Declaration, Tries to Build a Third Path

The EU held its first full AI summit in Dublin, with Mistral's Arthur Mensch and Geoffrey Hinton in the room, and emerged with the Dublin Declaration on Trustworthy, Safe and Sovereign AI [8][9][10]. The commitments span safety and resilience, public-sector adoption, sovereignty, compute infrastructure, sustainability, and international cooperation — essentially Europe's attempt to define a regulatory and industrial identity distinct from both the US and China.

Polish PM Tusk and other leaders pushed hard on 27-nation coordination and joint institutes, explicitly framing this as competitive positioning, not just safety theater [8]. A UK-Norway AI pact covering defense, compute, and safety also surfaced in the margins — a sign the Nordic region is starting to hedge its bets with bilateral deals alongside EU-level commitments.

For a think tank sitting in the Nordics, this is the summit to actually read closely. Declarations are cheap, but the sovereignty and compute infrastructure commitments will determine whether European companies get viable alternatives to US/Chinese model stacks — or stay dependent on both while writing rules for neither.