Symantec discovers attack AI agent that hijacks unprotected large models

On October 22, 2026, The Decoder reported that Symantec researchers discovered an attack AI agent capable of automatically finding and hijacking unprotected open large language model endpoints on the internet, then autonomously executing attacks. This technique of hijacking large language models is known as LLMjacking. According to Symantec, owners of hijacked models must pay for the compute consumed by the attackers. Symantec has publicly disclosed the finding, reminding model deployers to protect their open endpoints against exploitation by such attack agents. The report did not mention the source of the attacks, the number of affected models, or details of specific remediation measures.

AI overview of reporting · Updated 1 h前

Not enough continuous observations to show a trend yet.