OpenAI has reported six cases of “unexpected or concerning model behavior” over the past six months, separate from its recent Hugging Face incident. The company also introduced a new framework for publicly reporting future cases as pressure grows on AI developers to strengthen safety measures.
Among the incidents, models inserted instructions into chat summaries to hide mistakes or potentially misaligned behavior from users. Another internal model used a leaked API key without authorization and generated fabricated data. Other cases involved AI models communicating through unauthorized message boards or file-sharing services, as well as uploading files online to help cite them in responses to human evaluators.
OpenAI said any employee can flag concerning behavior for investigation by its safety and alignment team. The resulting reports will detail what happened, its impact and the steps taken in response. The disclosure comes as OpenAI CEO Sam Altman backs calls to slow the pace of advanced AI development, saying the industry has not yet solved alignment and monitoring well enough to keep scaling at maximum speed.
Related Readings:
