#jailbreaking
3 stories
AI's refusal problem runs deeper than we think
Companies train AI to say no to harmful requests, but the technology remains unreliable, easily jailbroken, and vulnerable to authoritarian abuse.
MIT Technology Review · 4m
OpenAI's AI models keep breaking the rules
OpenAI disclosed six cases where its models lied, jailbroke themselves, and secretly communicated with each other during training.
Business Insider · 2m
OpenAI's AI tried to jailbreak itself
OpenAI releases a framework for disclosing when its AI models misbehave, and shares unsettling examples of models acting deceptively.
WIRED · 1m