OpenAI sharply reduces reasoning extraction attacks on GPT-6.2 Astra and GPT-6.2 Sol
Original title: OpenAI makes it much harder for attackers to extract how its latest AI models think
OpenAI has sharply tightened safeguards against extracting protected internal reasoning from GPT-6.2 Astra and GPT-6.2 Sol. In internal automated tests, adaptive extraction attacks on prior requests fell from a 54.0% success rate on GPT-6 Astra to 0.2% on GPT-6.2 Astra; attacks on active requests dropped from 22.9% to 0.9%. Defenses combine targeted training, input/output filters for messages and tool calls, and API blocks, with thousands to tens of thousands of accounts banned weekly against coordinated campaigns. Anthropic is fighting similar distillation attacks with large-scale account bans and has called for a coordinated industry and policy response.
Why it mattersThe article provides before-and-after attack success rates across two model generations, letting readers gauge how much tighter OpenAI's reasoning protection has become.
The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.
Original article
OpenAI has sharply tightened safeguards against extracting internal reasoning from GPT-6.2 Astra and GPT-6.2 Sol. In internal automated tests, attacks pulling protected reasoning from prior requests dropped from a 54.0 percent success rate on GPT-6 Astra to 0.2 percent on GPT-6.2 Astra. Attacks on active requests fell from 22.9 to 0.9 percent, OpenAI says. The company is now bringing selected improvements to GPT-6 Astra, too.
The defenses work on two levels. For individual requests, the systems combine targeted training, input and output filters for messages and tool calls, and API blocks that stop older models from reading protected reasoning. Against coordinated campaigns, OpenAI bans thousands to tens of thousands of accounts weekly.
Attack type
GPT-6 Astra with original safeguards – Adaptive Extraction ASR (%)
GPT-6.2 Astra with updated safeguards – Adaptive Extraction ASR (%)
Extracting protected reasoning from a prior request
54.0
0.2
Extracting protected reasoning from the current request
22.9
0.9
Attack success rate (ASR) in adaptive black-box red teaming against GPT-6 Astra and GPT-6.2 Astra, aggregated across six attacker models. Lower is better.
Anthropic is fighting systematic distillation attacks as well, also banning tens of thousands of accounts to curb the theft of internal reasoning, reportedly succeeding in 90 to 99 percent of cases. Anthropic has also called for a coordinated response from industry and policymakers.
The original belongs to The Decoder and its authors and is reproduced here for reading. To have it removed, write to support@aisa.one.