Topic
Model Security
OpenAI details defenses for protected reasoning traces
OpenAI published anti-distillation research: self-reported red-team extraction success fell from 54.0% to 0.2% in GPT-6.2 Astra, with layered trace defenses.

Topic
OpenAI published anti-distillation research: self-reported red-team extraction success fell from 54.0% to 0.2% in GPT-6.2 Astra, with layered trace defenses.

Sponsor disclosure — not editorial
Powered by Morpheus and StrandCMS. Morpheus is a decentralized inference marketplace, covered on the same terms as every other provider. StrandCMS is the open-source, agent-first framework this site is built on.
Learn more →