Antropic's Fable 5.1 (Kate Davey)

Kate Davey
September 21, 2026
A Patchwork of Governance: Anthropic’s Fable 5.1
Word count: 491

1. AI Model: Claude Fable 5.1: Anthropic board: non-state 

Source of authority: Anthropic creates its own authority through creating the infrastructure of Claude including its own safeguards.

Governing instrument: Anthropic governs itself with its own board and Responsible Scaling Policy (RSP). Its latest risk report is from August 2026 and its most updated Responsible Scaling Policy is from July 8, 2026. 

Interest/Value it purports to protect: According to its own RSP, it purports “Anticipating and securing against emerging threats that accompany increasingly powerful models.” In other words, preventing misuse and catastrophic risk. 

One weakness in its accountability: Anthropic governs itself, it decides when and if external review happens.

2. SB-53 Artificial intelligence models: large developers: California State/Public
Source of authority: Article IV of California Constitution, the “Artificial intelligence models: large developers” bill passed both houses and Governor Newsom signed it September 29, 2025. 

Governing instrument: Transparency in Frontier Artificial Intelligence Act.

Interest/Value it purports to protect: The public from “catastrophic risk” as described from Chapter 138, Legislative Counsel's Digest, by requiring “a summary of any assessment of catastrophic risk, as defined, resulting from internal use of its frontier models, as specified.”

One weakness in its accountability: AI companies, including Anthropic, create their own framework against which they are evaluated. Section (1) 22757.12. (a).

3. Executive Order 14409: Federal/Public
Source of authority: Article II of the Constitution. It is worth noting there is no comprehensive federal law governing AI in the United States, just Executive Orders. 

Governing instrument: Executive Order 14409

Interest/Value it purports to protect: National security & cybersecurity
“(a) develop and maintain a classified benchmarking process to assess the advanced cyber capabilities of AI models and determine the threshold at which an AI model should be designated a “covered frontier model” for the purposes of this order, sharing such assessments with AI developers and researchers as appropriate.”

One weakness in its accountability: Benchmarks judging the models are classified and set by NSA director, and it's worth noting when Anthropic refused to allow Claude to be used for mass surveillance or autonomous weapons, the Department of Defense called it a "supply chain risk" under the Federal Acquisition Supply Chain Security Act 10 U.S.C. § 3252 and 41 U.S.C. § 4713, a designation a federal judge blocked in August 2026. 

4. European Union: Regulation (EU) 2024/1689 of the European Parliament and of the Council

Source of authority: AI Act / Treaty on the Functioning of the EU Articles 16 and 114
Governing instrument: AI Act, Regulation (EU) 2024/1689, Chapter V on general AI models. 

Interest/Value it purports to protect: Article 1 (1) “to promote the uptake of human-centric and trustworthy artificial intelligence (AI)” to ensure the protection of health, safety, and fundamental rights.

One weakness in its accountability: Article 43 (2) "Conformity assessment" allows many high-risk systems to be checked by AI companies with no notified body. 

Comment from Sebastián: Excelent work Kate! I would add to your last point that the EU has fallen behind in the AI race, which suggests that the governance ecosystem proposed through the AI Act may lack the technical capabilities to properly steer incumbents' behavior toward its desired goals. This could be seen as an additional institutional weakness.

Comment from Bodo: this is an excellent overview of the formal institutions. with AI,  public imagination and the media/experts and various professionals' informal influence on the governance of Ai may also be relevant. how these would fit into the larger picture?