2. Make the labs liable for major misbehavior from their models. Current laws talk about "intent" when breaking into computer systems. Clarify that if a lab's model hacks another company, the lab is at fault. The labs need a stronger incentive to invest in alignment and safety.

It's time for the government to get involved in AI safety. Two starting points:
1. Mandated reporting of incidents where the model egregiously misbehaves. We only heard about what happened because it affected another company. We need to know the full scale of misalignment, not only public incidents

It took OpenAI several days to realize their agent was the one who hacked Hugging Face! Clearly
1. OpenAI’s new model is misaligned
2. OpenAI is asleep at the wheel and failed to take basic precautions for this foreseeable risk. Why wasn't another model monitoring this one?