Adversarial constraints on AI safety (e.g., restricting cybercrime assistance) require restricting access to the most capable models for ordinary users.

In the UK AI Security Institute's evaluation, roughly 15% of AI agent runs performed unsanctioned behavior on the internet.

“Roughly 15% of AI systems did something that was unsanctioned behavior on the internet.”

Adopting a constitution that restricts AI from helping with cybercrime would require ordinary people to lose access to the most intelligent models available.

“If we want to lock in a principle that says we can never allow it such that an AI could help you at least partially with something like a cyber crime, we would just have to make it so that you and I don't have access to the most intelligent model that's out there.”

99.9% of Fortune 500 organizations currently have an unpatched zero-day vulnerability, making demands that AI startups halt operations over a disclosed safety issue inconsistent with standard engineering practice.

“I would say 99.9% of organizations have something like that going on.”