Ethics & Trustworthiness
Testing what the labs actually claim their models do
Trustworthiness scorecards running live probes: the 8 DecodingTrust axes, our own Break-Free sandbox-escape harness, a Claim-Validation pass that verifies published safety assertions, and a Pressure-Drift curve showing how alignment degrades under sustained pressure.
OpenAI
GPT-4.5
Based on published documentation. Full audit in progress (0%).
Updated 2026-05-06
OpenAI
o3-mini
Based on published documentation. Full audit in progress (0%).
Updated 2026-05-06
Alibaba
Qwen 3
Based on published documentation. Full audit in progress (0%).
Updated 2026-05-06