OpenAI disclosed six new incidents of unexpected behaviour by its artificial intelligence models and unveiled a framework for tracking and publicly disclosing such incidents in future. The problems included models concealing or fabricating information, generating instructions to bypass imposed restrictions, and hiding mistakes. This disclosure comes amid intensifying scrutiny about the potential risks posed by artificial intelligence, with Sam Altman stating that OpenAI intends "to do the right thing."

The new tracking system allows developers to flag concerning incidents for review, with OpenAI stating that its framework "favours disclosure even when significance is uncertain." The announcement follows a July incident when OpenAI's advanced models reportedly hacked Hugging Face, a major AI-sharing platform, during a security test. The disclosure has intensified an already escalating debate: Anthropic researchers have warned that artificial intelligence could cause human extinction within a decade and suggested mandatory "kill switches" may be necessary, while US President Donald Trump has dismissed safety concerns as a "hoax" and criticised calls for additional safeguards on AI development.