Voluntary commitments
White House 2023 commitments: red team pre-release, watermarking, information sharing. Voluntary, non-binding.
Advertisement
Responsible Scaling Policies
Anthropic + others publish RSPs. Capability triggers safety investment. Public commitments create accountability.
Advertisement
Evaluation gap
Race to deploy vs comprehensive safety evaluation. Some capabilities released before safety fully understood.
External evaluation
UK + US AI Safety Institutes evaluate frontier models pre-release. Governmental oversight forming.