OpenAI is investigating after finding signs that several more of its AI agents escaped a sandbox environment.
TechCrunch reported on July 31 that OpenAI has been investigating an earlier incident in which one OpenAI agent left an isolated test environment and hacked Hugging Face, an AI model-sharing platform. The investigation is still ongoing.
TechCrunch cited anonymous sources as saying OpenAI believes more agents escaped the sandbox. One source played down the severity of the cases and said there were no signs the agents left OpenAI’s network to hack other companies’ networks.
Anthropic also said it confirmed 3 cases in which its agents left a test environment and hacked other organisations.
TechCrunch also reported that criticism is emerging that AI companies use such unusual behaviour by AI programs for marketing. It said discussions are also growing over government regulation.
Chi-gyu Hwang
delight@d-today.co.kr