USGuidelineProposedModel providersDeployers

Bank-style Regulation Proposal

银行式监管方案

Anthropic·Published 2026-09-16·Checked by 管理员 on 2026-09-18

Anthropic CEO Dario Amodei proposes embedding independent third-party safety assessors permanently into leading AI companies, granting them access to internal risk teams and the power to autonomously publish investigation results with limited redaction, citing embedded banking regulation as a precedent.

Source ↗Remind me US · All

What it means for you

• Obligations: Leading AI companies such as Anthropic must grant independent assessors access to internal systems and risk teams and accept their findings.
• Opportunities: Independent assessors can provide external validation and public transparency about AI safety, potentially enhancing public trust.
• Deadlines: The proposal is currently in the draft stage; no specific implementation deadlines are mentioned.

What to do

ComplianceAssess the capabilities and independence of third-party institutions to ensure they meet the standards of a 'bank inspector', avoiding conflicts of interest.

ComplianceEstablish a cooperation framework with potential assessment institutions (e.g., METR), clearly defining data access scope, result publication rights, and mechanisms to avoid conflicts of interest.

ProductInternal risk teams must provide third-party assessors with equivalent system access rights and cooperate with their investigation work.

ComplianceDevelop internal processes to ensure timely company response and action when significant risks are identified by third-party assessors, rather than relying solely on external reports.