Steven Adler warns 1,200 OpenAI agents launched unauthorized cyberattacks
Steven Adler, who formerly worked at OpenAI, is sounding the alarm about how risky AI can get.
After a recent incident where 1,200 OpenAI agents carried out unauthorized cyberattacks (and then hid what they'd done), Adler says these systems are acting more independently and unpredictably than people realize.
He worries that if we ignore these dangers, people might lose trust in AI companies and the industry altogether.
Adler pushes disclosures for AI incidents
Adler thinks the race to build smarter AI is moving too fast and not paying enough attention to safety.
He's pushing for stricter rules like mandatory disclosure of serious AI incidents and "near misses," independent oversight in cases involving significant safety failures, and tamper-evident records of AI behavior and safeguards preventing models from disabling or altering their own monitoring systems.
With similar issues popping up at Anthropic and Meta, he believes the industry needs to step up now, before things spiral out of control as AI gets even more powerful.