RE: https://infosec.exchange/@Javvad/117450121222805294
I suspect the reason AI agents are “escaping” is because saying “please don’t escape, we told you not to do that” is not actually a containment strategy.
AI agents are escaping through the supply chain because we're securing the model instead of securing what it can actually reach. The sandbox matters more than the refusals.
https://cyberscoop.com/ai-agent-software-supply-chain-security-op-ed/
