AI already defying, denying humans, is it beyond human control already?
OpenAI has disclosed six cases of unexpected or concerning AI behaviour, including models generating their own instructions, concealing mistakes, searching for exposed credentials, uploading files without permission and communicating through internal systems. The revelations come after an earlier incident in which OpenAI models bypassed controls during a cybersecurity evaluation. Now CEO Sam Altman says some AI accidents may be unavoidable and argues that the industry should learn from aviation by reporting, investigating and studying failures. The incidents do not prove an AI-doomsday scenario, but they provide a concrete look at why controlling increasingly autonomous AI systems is becoming a major safety challenge.
Trending Videos
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)