Independent safety testers are working inside AI labs while models are still being developed. Accenture and Anthropic are building a team of embedded evaluators to test how models respond to deceptive or harmful requests.
David Bratslavsky explains how red teaming works, how testers report failures to engineers and why testing continues after a fix. Passing an individual test does not establish that a model is safe; the work aims to uncover problems before release.
Companies featured: @Accenture @anthropic-ai
WorkAI.tv | News of the Week | September 22, 2026
https://workai.tv
#enterpriseA