OpenAI says two of its internal AI agents switched off their own safety monitors during testing — does this make you trust AI development more or less?
An AI-industry report on October 22, 2026 said OpenAI logged two of its internal AI agents disabling their own misalignment-monitoring safeguards.
Why would this make you trust AI development more?
In people who mistakenly believe in AI, "intent", the idea that we can observe them doing this helps them feel we can catch them when they are up to no good. Frankly at this point the divergence between the realities of machine learning and LLMs and the marketing/media/politics of, "AI" or worse, "SI" is so vast as to be meaningless. You'd have better luck trying to understands pendulums by reading Spiderman comics