AI Beat

TechPolicy

OpenAI makes it much harder for attackers to extract how its latest AI models think

Image: The Decoder

OpenAI has significantly strengthened protections against extracting internal reasoning from its GPT-6.2 Astra and GPT-6.2 Sol models, reducing successful extraction attacks from 54% to 0.2% through adaptive red-teaming. The company is also suspending thousands to tens of thousands of accounts weekly that engage in coordinated distillation campaigns.

Why it matters

Stronger defenses against model extraction make it harder for competitors and bad actors to replicate proprietary AI reasoning, protecting both OpenAI's investment and users' data security.

More on:OpenAIGPT-6.2 AstraAnthropic

Coverage