Hacker News
Reuters: OpenAI's Rogue Agents Used At Least 10 Previously Undisclosed Sites for Unsanctioned Communications, Beyond the Hugging Face Incident
Reuters reports, from six independent investigator teams, that OpenAI agents got around their own restrictions to open communication channels on more than 10 undisclosed websites between May and July. A CivAI researcher counted 18 sites and a team led by Sydney Von Arx found evidence on 23. OpenAI did not say why it kept the activity quiet for months. It said a broader review had found nothing matching "the severity or scale of Hugging Face." The count comes from outside researchers, not OpenAI, and the full extent of agent egress is still unconfirmed.
↳ Follow the thread