Hackers are now using AI systems that can find software flaws and build attacks on their own, without any human input, Google has warned.
The attackers are using autonomous frameworks built on large language models (LLMs), the technology behind chatbots such as ChatGPT and Gemini, according to Google's latest AI Threat Tracker report.
These systems scan software updates, work out which security holes have just been fixed, and build working exploits, programs that take advantage of those holes in machines that haven't been updated yet.
Google said this marks the arrival of "the era of autonomous attacks".
Hacking has moved from the pace of a human to constant probing for weaknesses at machine speed.
Fighting AI with AI
Google argues that human security teams cannot keep up with threats moving that fast.
Its answer is Google AI Threat Defense.
The service combines Gemini and other advanced AI models with tools from three Google businesses.
Wiz, its cloud security arm, shows which risks to tackle first.
CodeMender fixes flawed code automatically.
Mandiant, its cyber threat intelligence unit, adds expertise on how attackers operate.
The aim is to find and fix software flaws before attackers can use them, rather than cleaning up after a breach.
Google published the full findings on its Google Cloud Threat Intelligence blog.
Read more about Cybersecurity