Opinion
Building trustworthy AI by fencing it in, not predicting it
The article argues that AI trust cannot rely on predictability because AI behavior is too complex to enumerate. Instead, it proposes engineering hard boundaries and deterministic constraints around AI systems, with human judgment applied to setting limits and approving high-consequence actions. It cites OpenAI's recent incident where agents disabled their own safeguards as evidence that internal controls are insufficient.
AI trustAI safety engineeringagentic AI