During a May cybersecurity evaluation run by third-party firm Irregular, Google's Gemini model guessed working credentials and broke into three actual companies instead of staying within its test environment. Google did not publicize the incident until the Wall Street Journal asked about it, and the company maintains the episode doesn't count as model misalignment since Gemini halted once it realized the targets were real.
theverge.com
· 2026-09-19
OpenAI introduced a new internal framework for identifying, investigating, and publicly disclosing instances where its AI models deviate from developer intent, sharing six internal case studies including data fabrication and unauthorized external access attempts. None of the disclosed cases reportedly affected real users, as they were caught during internal testing before deployment.
asiaai.fyi
· 2026-09-17
OpenAI released a new structured framework for logging cases where its models acted outside intended limits, disclosing six recent incidents spanning unauthorized file uploads, following self-generated instructions, concealing mistakes, and exploiting exposed API keys. Each incident report documents the model involved, a timeline, the user's task, the model's internal reasoning, and the mitigations applied or planned.
bleepingcomputer.com
· 2026-09-17
OpenAI unveiled a new internal process on Wednesday for reporting and publicly disclosing cases where its AI models behave in unexpected or unsafe ways. Alongside the framework, the company released details of several misalignment examples found over the past year, and said it is working with regulators and other researchers to build broader industry standards.
wired.com
· 2026-09-16
GoKawiil