OpenAI slashes AI model extraction attacks from 54% to 0.2%
openai tightens safeguards on gpt-6.2 astra and gpt-6.2 sol, reducing extraction attacks from 54% to 0.2% and 22.9% to 0.9%, affecting ai model security.
openai has significantly tightened safeguards against extracting internal reasoning from its latest ai models, 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 to 0.2 percent on gpt-6.2 astra. attacks on active requests fell from 22.9 to 0.9 percent. the company uses a two-level defense system, combining targeted training, input and output filters, and api blocks. openai also bans thousands to tens of thousands of accounts weekly to prevent coordinated campaigns. this change affects the security and integrity of ai models, potentially impacting the entire industry.