The system, which GTIG calls exploit_agent, compares code before and after a patch, identifies likely flaws, writes exploits and runs them automatically, the report said.[1] This is the first time Google has observed such a fully autonomous attack pipeline in use.[1]
Patched flaws may be easier targets for AI than unknown ones, Google said.[1]
In tests, only three of seven AI models found the bug, with GPT 6 Sol identifying it in 13 minutes for $4.24 by September, ZDNET reported.[2] The falling cost of token usage has made such attacks cheaper to run, ZDNET added.[2]
Several groups now use open-source AI hacking frameworks in live attacks, GTIG said, including financially motivated gangs and suspected Chinese spy groups.[1] A configuration file on the server carried comments in Russian, and victim data suggested targeting of French media, entertainment and education websites, GTIG said.[1]