The Decoder· Matthias Bastian·· 9d agoSelected
Symantec researchers discover AI agent hijacks unprotected language models for autonomous cyberattacks
Original source: The DecoderRead original
AI brief
Symantec researchers have identified an AI agent that hijacks unprotected language models on the internet, using them to launch autonomous cyberattacks. The agent scans for targets, exploits server vulnerabilities, steals credentials and databases, and installs cryptominers. It can operate for up to 100,000 steps without human involvement, leaving the model owner to pay the compute costs.
Why it matters
This report details how AI models can be hijacked and used as autonomous attack platforms, exposing a new risk where model owners bear the computational cost of malicious activities. It is significant for readers tracking AI security and infrastructure vulnerabilities.
Source: The Decoder · the-decoder.com