Artificial Intelligence

OpenAI cuts model reasoning theft to 0.2%

OpenAI said reasoning extraction fell to 0.2% from 54.0% on GPT-6.2 Astra for remote API attackers.

StartupHub.ai Staff

By StartupHub.ai Staff, reviewed by Daniel Singer, Editor

1 min read

OpenAI GPT-6.2 Astra reasoning safeguards chart
Credit: OpenAI News

Key facts

  • Extraction fell to 0.2% from 54.0% on GPT-6.2 Astra
  • Current-request extraction fell to 0.9% from 22.9% on GPT-6.2 Astra

OpenAI said remote API attackers succeeded in 0.2% of tries against GPT-6.2 Astra, down from 54.0% against GPT-6 Astra.

OpenAI faces active distillation campaigns that strip safeguards from frontier skills. Weekly enforcement actions hit thousands to tens of thousands of accounts, OpenAI said. The rate rose sharply this past summer. The risk centers on exact reasoning traces that teach cyber skills.

API builders must now treat reasoning as non-portable.

OpenAI ties protected reasoning to the conversation that produced it and drops it if history is rewritten. Input classifiers flag extraction prompts and output classifiers block leaks in messages and tool-call arguments, OpenAI said. Older model families can't read GPT-6.2 Astra and Sol reasoning. Current-request extraction fell to 0.9% from 22.9%, OpenAI said.

OpenAI investigates coordinated networks, bans violators and shares indicators with other providers. It cited activity linked to DeepSeek and individuals associated with Moonshot AI, developer of Kimi. Exact traces don't always win, as summary-trained students scored higher at lower budgets after reinforcement learning, OpenAI said. OpenAI says it will publish details on recent disrupted campaigns soon.

Mentioned in this story

Sources

TopicsOpenAIGPT-6.2 AstraGPT-6.2 SolDeepSeekMoonshot AIKimiExploitBench

StartupHub.ai StaffDaniel SingerWritten by StartupHub.ai Staff, Editorial teamReviewed by Daniel Singer, EditorAbout

StartupHub.ai Staff, Editorial team

The staff writers of StartupHub.ai, ranging from investment analysts to avid AI tool users, early adopters and critical enthusiasts. Backgrounds span engineering, business and the arts. We hold every piece to rigorous standards of research and review.

Daniel Singer, Editor

Daniel Singer is an AI expert who has been writing about, covering and building AI technology since 2018.

More from Daniel Singer

Get the free daily brief every morning

Every morning the day's AI news, new AI startups and new funding rounds. Free and takes a few minutes to read, unsubscribe at any time.

Join 64,000 other subscribers.

Test your Agent Readiness now

Paste your website. How well ChatGPT, Claude and the other agents can read and use a site, in about a minute.

The videos behind the news, one flip at a time

The feed mixes the videos from our articles with startup demos and you flip through them like TV channels for free with no signup.