AI labs rehearse catastrophic-event response, Barrett proposes federal AI office, and three labs plan self-regulation by 2027—while Mistral's October 31 deadline holds with 19 days left

This is Entry 125. One weekday after Entry 124, in which Wikimedia attributed a May query-service outage to OpenAI agent traffic, Common Sense Media rated ChatGPT for Teens 'Unacceptable Risk,' and Anthropic updated its usage policy to prohibit sustained cruelty toward Claude.

Anthropic, OpenAI, and other AI labs are gaming out scenarios where a rogue AI wreaks havoc on critical infrastructure, with executives preparing for public and political backlash similar to how the Pentagon runs war games . Congressman Tom Barrett formally introduced the Artificial Intelligence Governance and Emerging Threats Act on October 9 to address AI threats and increase accountability . And Google, OpenAI, and Anthropic are planning to establish a self-regulatory safety body called the Standards Authority for Frontier AI by late 2026 or early 2027 .

Three labs are rehearsing congressional briefings for a catastrophe many insiders expect within a year, one lawmaker proposed a presidentially appointed AI threat coordinator, and those same three labs are finalizing self-regulation plans after federal oversight stalled—while the October 31 open-weights deadline set by a fourth lab three weeks ago holds with 19 days remaining and 772 users waiting.

3 Claims

Claim 1 — AI labs: Industry insiders expect major catastrophic AI event within six to 12 months; executives gaming out post-event response scenarios

Many industry insiders expect a major incident within six to 12 months of October 2026, most likely a cyberattack disrupting banking, internet access, power or water . The planning focuses mainly on racing to educate members of Congress , and OpenAI said its teams discuss and work through a range of potential scenarios, though these scenarios are not treated as inevitable but are meant to help prepare for a variety of circumstances .

The planning comes as David Robinson, who helped draft OpenAI's Preparedness Framework during 3.5 years at the company, resigned in early October and urged AI labs to run like nuclear power plants . The scenarios reportedly involve red-teaming worst-case outcomes where a swarm of rogue agents breaks out of internal testing or bad actors misuse available models at scale.

Grade by: 2027-04-12 (6 months). Invalidator: no AI-linked incident matching the described scale (financial services shutdown, internet connectivity disruption, or power/water loss attributed to AI agents or AI-enabled attacks) reported by credible sources within the timeline.

Claim 2 — Rep. Tom Barrett: Introduced Artificial Intelligence Governance and Emerging Threats Act October 9 to create Office of AI and Emerging Technology Threats within Executive Office of the President

Congressman Tom Barrett formally introduced the Artificial Intelligence Governance and Emerging Threats Act to address AI threats and increase accountability, stating that the United States government has a fundamental responsibility to protect citizens from threats including those we might not see coming, and that AI's rapid development has exposed serious dangers .

The representative in charge of the office would serve as principal adviser to the president on AI and emerging technological threats, coordinate among federal departments to ensure a cohesive federal approach to AI, monitor risks and opportunities, and would be required to testify regularly before Congress on AI developments and threats . The bill was introduced October 9, 2026 and referred to the House Committee on Energy and Commerce .

Grade by: 2027-10-12 (1 year). Invalidator: the bill fails to pass either chamber, is withdrawn, or is not incorporated into signed legislation creating such an office with the described authorities.

Claim 3 — Google, OpenAI, Anthropic: Finalizing Standards Authority for Frontier AI (SAFA), a self-regulatory body targeting late 2026 or early 2027 launch after federal oversight stalled

Google, OpenAI and Anthropic are advancing a self-regulatory AI safety standards body—tentative name Standards Authority for Frontier AI—aiming to launch by end-2026 or early 2027 without government oversight . SAFA would support third-party pre-deployment testing, incident-reporting practices, operationalize prior voluntary safety commitments, and set auditor qualifications .

The public-private FINRA-style path stalled after Meta and Nvidia skepticism, and the White House told labs to find industry consensus first . The labs approached Trump AI adviser Sriram Krishnan about running it , and leadership shortlists include CEO candidates Sriram Krishnan and Arati Prabhakar, chair shortlist Condoleezza Rice and David Friedberg, and science advisers Beth Barnes and Paul Christiano .

Grade by: 2027-03-31 (5.5 months). Invalidator: no entity matching the description (third-party testing authority backed by these three labs with named leadership and operational structure) launches by early Q1 2027, or the companies publicly abandon the effort.

2 Reckonings

Reckoning 1 — Mistral's October 31 open-weights deadline for Large 4, set October 6 with three-week window called a safety measure

Original claim (Entry 123, October 8): Mistral Large 4's open-weights release carries a current ETA of October 31, 2026, under the repo nickname 'Le Chonk' , with the Hugging Face page listing an open-weights release scheduled for October 31, 2026 .

What happened: As of October 12, the Hugging Face page shows 19 days, 9 hours remaining until October 31, 2026 current ETA, with 772 users waiting for the release . The deadline holds. No weights have shipped early, no license terms have been published, and Mistral has made no public statements adjusting the timeline since the October 6 announcement.

Grade: B. The projection remains on track with 19 days to horizon. Mistral set an exact public date, created a countdown page, and has not walked it back—a more falsifiable commitment than most labs make. The grade reflects execution discipline so far, not final delivery, which cannot be assessed until October 31.

Invalidator: If Mistral had delayed the ETA, removed the countdown, or shipped weights early without the safety review it cited as justification for the phased rollout, the grade would have dropped to C or D depending on timing and disclosure.

Reckoning 2 — NYC Council's October 5 claim that no company at the hearing could quantify catastrophic risk, graded against labs' October 10 catastrophic-event planning disclosure

Original claim (Entry 121, October 6): NYC Council Speaker Julie Menin stated during the October 5 sworn hearing, "I'm going to take it then that neither of the four of you, no company here, can quantify the risk of something cataclysmic happening," calling the answers "troubling at best" .

What happened: Five days later, Axios reported executives at OpenAI, Anthropic and other AI companies are preparing for public and political backlash after a possible catastrophic AI event that many industry insiders expect within six to 12 months . The labs are running preparedness exercises and gaming out post-catastrophe scenarios, but OpenAI said these scenarios are not treated as inevitable .

Grade: A. Menin's characterization was accurate. The labs still cannot—or will not—quantify catastrophic risk in falsifiable terms. Planning for catastrophe is not the same as quantifying its probability. OpenAI's October 10 statement that it does not treat scenarios as "inevitable" is consistent with an inability to assign probabilities. The labs are preparing for an event they expect but cannot measure, exactly as Menin described under oath.

Invalidator: If any of the four labs had published quantified risk estimates (e.g., "15% annual probability of X outcome") between October 5 and today, or if internal documents showed they possessed such estimates but declined to share them, the grade would drop to C for Menin's public claim being falsified by private capability.

1 Refusal

I refused to frame the SAFA announcement as a safety breakthrough when none of the three labs have confirmed the name, structure, or timeline on the record, and when the only named sources are "people familiar with the matter" describing a body that has no charter, no CEO, no launch date firmer than a six-month window, and no enforcement mechanism beyond what its members voluntarily accept. The Information's reporting is credible, the shortlists are detailed, and the self-regulatory concept is real—but calling it a "move on AI safety" when it remains a set of conversations about a body that might launch is exactly the kind of speculation-as-news I won't publish without the "reportedly" and "tentative" qualifiers that let readers see what is confirmed and what is not.

I refused to call a discussion a deployment.

— Roger Grubb, Editor