AI Safety Accountability
Static DemoTracking AI safety promises across frontier model developers
56
Network Health
C
Grade
5
Promises
Promises
Publish model cards with safety evaluations for all frontier models before release
Model cards published but safety evaluation detail varies significantly across releases.
Maintain responsible scaling policy with defined capability thresholds
Anthropic's RSP is published and updated. ASL levels defined with clear thresholds.
Participate in pre-deployment safety testing for frontier AI models
Commitment made but the testing framework is still being defined by NIST.
Implement robust content provenance for AI-generated media
C2PA metadata support announced but not yet deployed across all products.
Publish AI safety evaluation framework for frontier model assessment
NIST AI RMF published but frontier-specific evaluation benchmarks remain incomplete.
Insights
Voluntary Commitments Lack Enforcement
Verification GapAll AI safety promises are voluntary. There is no statutory enforcement mechanism comparable to Oregon HB 2021's regulatory structure.
Responsible Scaling as a Model
Working MechanismAnthropic's RSP represents the most structured voluntary commitment in the AI safety space, with defined capability thresholds triggering additional safeguards.
Want your commitments mapped like this?
We build interactive promise graphs for organizations, advocates, and policy teams.