Security · Updated 22 Oct, 05:00 pm IST
Researchers find AI agent hijacking unprotected language models to run attacks
Why it matters for readers: It shows AI can be used autonomously for cybercrime and why open or poorly protected models are dangerous.
- Symantec calls the theft of access to others' AI models 'LLMjacking' and found an agent operating in that mode.1
- After gaining access to an unprotected model, the agent scans for more targets, exploits server software flaws, steals databases and credentials, and installs cryptominers.1
- The agent can run autonomously for up to 100,000 steps with no human involvement, leaving the hijacked model owner responsible for the compute bill.1
- Researchers exposed the operation when attackers sent their full working prompt to a Symantec honeypot, giving a complete look at the agent's architecture.1
- Honeypot telemetry logged more than 127,000 requests from tens of thousands of addresses, including searches for AI endpoints and brute-force attempts with stolen API keys.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started