Scott Wiener, Author of California's SB 1047 and SB 53, Wins Nancy Pelosi's House Seat and Pledges Federal AI Oversight

This is a policy story, not a model launch. The METR assessment covered Anthropic's internal models and R&D pipeline, from pretraining to reinforcement learning (METR thread). On Nov 9 Anthropic released Claude Mythos 5.2 and Fable 5.2, and cut Fable's input price to $8 per million tokens (TechCrunch). Wiener's federal agenda includes:

• a federal AI regulator
• whistleblower protections
• testing and audit requirements
• a kill-switch requirement
• oversight of AI labs
• no broad federal preemption of state laws

(MIT Technology Review)

• Metric: AECI is Anthropic's version of the Epoch Capabilities Index. It combines benchmark scores through a logistic link, and METR fit trend breaks with ordinary least squares. Since October 2025, growth is 1.9x the earlier rate (95% CI 1.5–2.8) (METR).
• Attribution: METR's subjective guess is that about 80% of the acceleration comes from AI use in R&D. Staff disagreed on this. Employees self-reported 12x uplift on the newest internal model (METR).
• Mechanism: Coverage describes the gains as AI helping human R&D workflows, not fully autonomous research (Winzheng).
• Reproducibility: Low. Anthropic kept unilateral redaction rights and withheld most of the evidence METR considered most informative (METR).

• "Evidence of recursive self-improvement" (Wiener, many commentators): This is an interpretation. METR's measured claim is AI-assisted acceleration. It also warns that, because it found no clear bottlenecks, progress could speed up further (METR; Zvi Mowshowitz).
• Funding: Anthropic has given Public First Action $40 million. An Anthropic spokesperson said the money cannot be used to influence candidate elections (Mission Local). How that separation worked in practice is not independently documented.
• Margin: Wiener led 54.6% to 45.4% when the race was called (Mission Local).

• SB 1047 imposed duty-of-care obligations, and Newsom vetoed it. SB 53 replaced them with a "show your work" disclosure regime (StateScoop).
• Reps. Bill Foster and Ted Lieu pitched a House Select Committee on AI, with subpoena power, to Hakeem Jeffries (Nextgov). Wiener says he wants a seat on it.
• Ro Khanna has proposed an FDA-style AI agency with auditors embedded in frontier labs (Political.org). He has also proposed a ban on recursively self-improving AI until federal safeguards exist (Common Dreams).

• METR says the acceleration could be overstated if benchmarks are saturating or non-AI factors drive the trend. It could be understated if AECI misses real-world impact (METR).
• Anthropic says Mythos 5.2 and Fable 5.2 suspect they are being evaluated more often than any of its recent models. That makes it harder to predict how they will behave in real use (TechCrunch).
• Wiener reads the data-center backlash as driven by local costs like electricity bills and water, not by AI risk (MIT Technology Review).
• Opponents say the Anthropic-linked money puts Wiener too close to the industry. He calls that "shockingly cynical."

• Frontier labs: Expect pressure for SB 53-style disclosure, third-party audits and whistleblower rules at the federal level. Wiener opposes preemption, so a single national rulebook that overrides state laws looks less likely.
• Evaluators: The METR–Anthropic arrangement, with high access but lab-controlled redaction, may become the template for oversight. Its transparency gaps are already part of the debate.
• Infrastructure developers: Local opposition to data centers is a separate and immediate siting risk (Baxtel).
• Overall: The White House's dismissive stance makes near-term federal law uncertain.