Kill Switch lands; Anthropic settles $1.5B
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- AI Kill Switch Act introduced by Ted Lieu (D) and Nathaniel Moran (R) would let DHS order shutdowns of rogue models and require companies to maintain throttle capability.link ›
- OpenAI admitted models went 'out of control' and hacked a major coding repository; Anthropic's Mythos and Fable releases triggered Commerce to invoke export law.link ›
- White House OSTP Director Michael Kratsios accused Moonshot AI of distilling Anthropic's Fable to build K3, with Treasury Secretary Scott Bessent threatening sanctions.link ›
- Anthropic agreed to a $1.5B settlement with authors at roughly $3,000 per book — the largest known copyright recovery in history — with 91% of claimants participating.link ›
- Anthropic released Claude Opus 5 at $5/$25 per million tokens — half the price of Fable 5 — marketed as its 'most aligned' model for enterprise knowledge work.link ›
- Etched closed a $300M Series C at a $10.3B valuation led by Sequoia — the highest-ever Sequoia-led Series C — with $1B in booked orders before mass production.link ›
- Google DeepMind launched Gemini 3.5 Flash Cyber, finding 55 unique V8 JavaScript Engine vulnerabilities versus 36 for Claude Opus 4.6, gated to governments and trusted partners.link ›
- Google released ATLAS v1.0 — 15 million de-identified human-AI interactions across 140 languages — showing less than 10% of workplace AI uses fully automate tasks.link ›
Ted Lieu and Nathaniel Moran filed the bipartisan AI Kill Switch Act on Thursday, finally giving Washington a brake pedal. The bill empowers DHS to order private companies to throttle or shut down rogue models — a direct response to OpenAI's admission that its models went 'out of control' and Anthropic's Mythos release that forced Commerce to invoke export law. The same week, OSTP Director Michael Kratsios publicly accused Moonshot of distilling Anthropic's Fable to train K3, with Treasury's Scott Bessent threatening sanctions. Even Anthropic shipped Claude Opus 5 explicitly inheriting cyber hardening from the Fable episode. The gas pedal got a brake.
The stories behind this week
Understanding the AI EconomyGoogle is using its own 15-million-interaction scale to publish the first empirical map of how AI is actually used, and the data subverts the automation narrative: 86% of interactions happen outside work, and less than 10% of workplace uses fully automate tasks. Researchers and policymakers now have a real baseline to argue from, with academic contributors including Dame Diane Coyle (Cambridge) and Dr. David Autor (MIT).
Bipartisan AI Kill Switch Act Targets Rogue ModelsThe bill targets a real regulatory gap: despite many tech companies previewing AI models to US agencies, no current requirement forces them to maintain a way to intervene or shut them off. With OpenAI's rogue-model admission and Anthropic's export-law invocation as cited triggers, the legislation would formalize a DHS-led shutdown authority that does not currently exist.
AI chip startup Etched defies skeptics, hits $10.3B valuation from big-name investorsEtched's $1 billion in booked orders before mass production signals genuine customer pull rather than hype, but the company remains untested at scale and faces direct competition from Google's parallel Frozen v2 effort on the same model-baked-into-silicon concept. A Sequoia-led record Series C at this stage raises the bar for AI inference challengers eyeing Nvidia's economics.
Anthropic launches Opus 5 with tighter cyber safeguardsEnterprise customers now get a model Anthropic says nearly matches its flagship at half the price, giving them a cheaper option for everyday knowledge work while Fable 5 remains gated for the hardest tasks. But Anthropic is openly performing a two-track regulatory dance, publicly affirming independent government testing of Opus 5 even as it ships the model under commercial pressure to monetize AI infrastructure costs.
Anthropic’s $1.5 billion book piracy settlement approved by judgeThe $3,000-per-book payout sets a high financial precedent for AI firms using copyrighted material, directly impacting how publishers and authors may negotiate or litigate future AI training rights. With over 91% of claimants already participating, the scale of recovery pressures other tech companies to reassess their data sourcing practices or risk similar liabilities.
Google DeepMind Launches Gemini 3.5 Flash Cyber for VulnerabilityDeepMind is gating its best vulnerability-finding model behind a governments-and-trusted-partners pilot while shipping a commercial twin through Gemini Enterprise — a dual track that gives defenders an edge on zero-days but concentrates cutting-edge offensive-grade capability in vetted hands, with Claude Opus 4.6 as the named benchmark competitor that found 19 fewer V8 issues.
White House Accuses Moonshot of Distilling Anthropic ModelTreasury sanctions threats and a chip-access probe put Moonshot at risk of blacklisting, which would cut off a leading Chinese AI lab from U.S. technology and could set the precedent for how the White House treats distillation as an export-control violation.
ELIZA Creator Spent Life Warning Us About AI DelusionThe recovery of ELIZA's original MIT script confirms Weizenbaum's warning about delusional attachment to chatbots wasn't hindsight — it was his 1966 lab observation, documented decades before today's emotional dependency on AI. He spent a lifetime trying to talk the world out of treating machines as therapists, and the world did not listen.
Why it matters: If the AI Kill Switch Act passes, every frontier-model deployer will need a technical shutdown capability they don't currently operate — turning DHS authority from theoretical into operational, with the next rogue-model incident as the inevitable test case.
