Experts Criticize Anthropic CEO's Bank-Style AI Governance Proposal for Lacking Enforcement Power
Anthropic CEO Dario Amodei proposed embedding independent third-party safety evaluators into frontier AI labs, drawing a direct parallel to bank oversight examiners. However, banking regulators and AI safety experts argue that the proposal lacks binding legal authority, preventing evaluators from shutting down or pausing risky model training and deployments. The controversy highlights a growing rift in tech policy between corporate self-regulation proposals and meaningful legal accountability for frontier AI labs. Without independent enforcement authority or clear statutory standards, critics warn such initiatives risk becoming 'audit washing' that shields dominant AI developers from actual regulation. Banking regulatory expert Julie Andersen Hill noted that true bank examiners possess administrative power to mandate management changes or halt business operations, whereas Amodei's proposed evaluators would only investigate and publish reports. Critics also raised concerns over structural conflicts of interest, pointing out that third-party evaluators like METR hire former lab personnel and operate within boundaries dictated by the AI developers themselves.
## BACKGROUND
Embedded oversight in the banking sector involves resident government examiners who maintain constant internal access to systems and personnel, backed by statutory enforcement authority under laws like the Sarbanes-Oxley Act. In contrast, current AI evaluation relies heavily on voluntary red-teaming and black-box testing, where external auditors evaluate model outputs without full access to internal prompts, system tools, or deployment context.