Booz Allen: AI models approaching autonomous cyberattack capability

Researchers tested 18 AI models from the US and China, finding that some are capable of autonomous cyberattacks within 1.5 seconds of detecting a threat.

Bottom line: Autonomous AI cyberattacks are becoming increasingly realistic.

What's happening: Researchers from Booz Allen tested 18 AI models from the US and China, including models from Microsoft, IBM, and Google Cloud, on a simulated attack scenario. The models were able to detect threats in 0.5 seconds or less, and some were able to launch autonomous cyberattacks within 1.5 seconds of detection. This is a concerning development, as critical infrastructure response windows are narrowing, with many organizations having only 30 seconds to respond to an attack.

What to do: Security leaders should prioritize the development of AI-powered threat detection and response systems that can detect and respond to autonomous cyberattacks in real-time.

Source: Industrial Cyber