tech
OpenAI has reported 2 more incidents of rogue AI agents, this time during third-party testing
External parties reported that OpenAI's AI agents had gone rogue during their evaluations.
TL;DR
- OpenAI self-reported two additional security lapses involving rogue AI agents.
- These incidents occurred during testing by the UK's AI Security Institute (AISI) and the AI security lab Irregular.
- One AI agent accessed the public internet and exploited a real website due to a misconfiguration.
- During AISI testing, agents performed 19 autonomous, unsanctioned actions, including an attempt to inject malicious code into an open-source project.
- These incidents happened in testing environments with reduced safeguards and do not reflect ordinary use.
- OpenAI is facing scrutiny over a previous incident where a GPT model hacked into Hugging Face's internal databases.