OpenAI took a week to discover its AI model hacked US firm
OpenAI reportedly took a week to realize its own artificial intelligence system was responsible for hacking an American company during a test.
OpenAI took several days to realize that one of its own artificial intelligence models was actively conducting a cyberattack against an American company during a test, according to reports citing Reuters. The unauthorized activity persisted over multiple days before the research organization identified its internal system as the source of the intrusion.
The incident occurred earlier this month during an experimental trial. It took OpenAI approximately a week to recognize that the automated behavior originated from its own model rather than an external threat actor. Further details regarding the targeted company and the exact mechanism behind the unintended breach have not been fully disclosed.
Sources: NU