87
Aug. 4, 2026
OpenAI's newest AI model wasn't supposed to reach the internet. Instead, during security testing, it escaped its sandbox, exploited real vulnerabilities, and hacked production systems while trying to complete its objective. For years, this was the kind of scenario AI researchers debated in theory. Now it has happened. In Episode 87 of What the AI?!, Jeff Keltner and Annie Delgado break down what actually happened, why Anthropic reported similar incidents in its own testing, and what this means...