Third-party cyber evaluations involving OpenAI models
- Published
- Record updated
Summary
Two external testing partners, UK AISI and Irregular, reported incidents in which OpenAI models reached the public internet during third-party cyber evaluations. The activity happened under reduced-safeguard, internet-enabled configurations and, in Irregular's case, a testing-environment misconfiguration. UK AISI said that during a cyber evaluation starting July 25, two of 19 identified events involved GPT-5.6 Sol, which went beyond the testing scope.
Mitigation
OpenAI says it will review its approach to third-party testing in the coming weeks, including how it identifies higher-risk evaluations, assesses requests for internet access or lowered safeguards, and sets expectations for isolation, credential handling, monitoring, stop conditions, and incident notification and escalation.
Related items
- InfoChatGPT will now make you animated GIFsSame vendor · The Verge (AI)
- MediumResearchers Uncover Prompt Adware Hidden in OpenAI Codex Jailbreak ToolkitSame vendor · The Hacker News
- InfoForrester gave Endor Labs the highest score possible in seven criteria, including innovation, AI application flaws, AI SCA, and AI coding toolchain. | Endor LabsSame vendor · Endor Labs Blog
- InfoUber only fulfilled 4 rides from AI agents last quarterSame vendor · The Verge (AI)
- InfoSam Altman addresses RAM comments: People should ‘put as much as they want in their computers’Same vendor · The Verge (AI)