Charting a path to AI accountability

Summary

Anthropic’s response to the NTIA’s Request for Comment on AI Accountability proposes processes and infrastructure for evaluating highly capable and general-purpose AI models. It recommends funding rigorous evaluations, requiring near-term disclosure of evaluation methods and results, and developing longer-term industry standards with agencies such as NIST. The framework would use capability-based risk thresholds, pre-registration of large training runs, confidential registries, technically literate and security-conscious third-party auditors, and external red teaming before advanced model release. It also calls for more interpretability research, acknowledges that regulations requiring interpretable models are currently infeasible, and seeks antitrust guidance for industry safety collaboration. Anthropic says these measures require collaboration among researchers, AI labs, regulators, auditors, and other stakeholders to build accountability and auditing mechanisms that support safe development and deployment.

Takeaways

1. 1

   Evaluation requirements would become risk-responsive: models below an established threshold could follow existing safety standards, while models exceeding it would face stronger oversight or halted deployment if assessments and mitigations are insufficient.

2. 2

   Large AI training runs would be pre-registered confidentially with national governments, including model specifications, compute infrastructure, intended completion dates, and safety plans, with aggregated data protected to high standards.

3. 3

   External red teaming is proposed as a precondition for releasing advanced AI systems, but high-quality external options need to be developed because red-teaming expertise currently resides almost exclusively within private AI labs.