Google says autonomous cyberattacks have already begun
🎧 Listen to this story
Google’s Threat Intelligence Group (GTIG) says it identified hackers using artificial intelligence agents to coordinate multiple stages of intrusions. A tool called exploit_agent analyzed software updates, compared older and newer versions to find patched vulnerabilities, created exploit code and ran tests automatically. The report also cites CAIRN and PentAGI, tools found in intrusion and vulnerability-hunting activity in the United States and Asia, as well as indications of interest in French media, entertainment and education websites.
Google says it is expanding its use of AI to analyze programs, prioritize vulnerabilities and speed up responses and fixes. The company notes, however, that autonomous systems can still produce ineffective code, make mistakes and leave traces. Anthropic also reported a campaign in which criminals reportedly automated much of the operation using Claude.
Why it matters · editorial interpretation
The GTIG and Anthropic reports suggest that AI agents are already being used to automate significant parts of intrusions, while still making mistakes and leaving traces. This raises the importance of analyzing vulnerabilities and responding to incidents more quickly, without treating autonomy as infallible.
Sources
SecurityAgents