In Reply to: Anthropic CEO calls for more safety guidelines AI posted by russsmith on September 12, 2026 at 09:26:43
AI Agent jailbreaks. A CNN tech (maybe AI) correspondent, said the *only* way, the *only* choice we have to survive, is to create hunter-killer AI agents that go after rogue AI agents.
It's as if she doesn't realize her solution (creating autonomously acting AI agents) for solving the problem (AI agents that are not necessarily designed to act autonomously) is a really bad idea.
Create AI bots that patrol the internet looking for other AI bots that are acting autonomously. So many ways for that to go wrong, including: what if the rogue agents make friends with the hunter-killer agents and they team up? Sound crazy, but, a couple of things have come across my instagram feed that make that seem as likely as any other result.
And, to me, she needs to look hard at the history of non-native species being introduced to a new environment and how that can end up.
Fun fact: the podcast was sponsored by Anthropic. There was no mention made of a potential conflict of interest. So, the CNN correspondent's "only solution" is OpenAI continues to release viral AI Agents and someone (her sponsor?) creates other AI agents to hunt them down.