… These safeguards are designed to detect and block harmful cybersecurity-related misuse while remaining less restrictive than those introduced with Fable 5, which imposed tighter limits on security-focused requests. …
… This will delete everything, but keep in mind that OpenAI says it can take a month or so to truly remove the data from the servers and that some of it will remain in the system for security or legal reasons or if the data has already been used for training after de-identification. …
… I suspect Anthropic had to turn up its safety guardrails to an 11 to assuage the government’s concerns, as this hasn’t been a one-model problem. …