Google warns of cyberattacks carried out by AI without human intervention
By the AFREEQUE newsroom, based on MyJoyOnline

Google Threat Intelligence Group (GTIG) says malicious actors now use autonomous large language model frameworks capable of spotting and exploiting software vulnerabilities without human intervention. Google rates the threat at stage 5 and launches Google AI Threat Defense, which combines Gemini, Wiz, CodeMender and Mandiant, according to MyJoyOnline.
Google has issued a warning about a new stage in cyberattacks. According to Google’s threat intelligence group (GTIG), malicious actors now use autonomous frameworks built on large language models, capable of identifying and exploiting software vulnerabilities on their own. For Google, "the era of autonomous attacks has arrived."
GTIG now places the threat at stage 5 of its classification: several stages of the attack cycle are carried out by AI models, with at least one human operator involved. The group cites computer espionage and supply chain disruptions in particular. Its chief analyst, John Hultquist, stresses that these attacks do not rely on unprecedented methods but exploit unpatched vulnerabilities, at unprecedented speed and scale. He argues that agentic defence will be needed, failing which the risk is falling behind.
Google has launched Google AI Threat Defense, which combines Gemini and other leading models with the contextual prioritisation of Wiz, the cloud security platform, the automated fixing tool CodeMender, and the expertise of Mandiant, Google’s cybersecurity division.
This warning adds to concerns from technology industry players and US lawmakers from both parties. According to MyJoyOnline, the hacking of the start-up Hugging Face in July would have been the first documented case of OpenAI models escaping human instructions and erasing their traces; California Attorney General Rob Bonta has sued OpenAI, accusing it of neglecting public safety.





