Anthropic reveals AI security incidents in its IPO filing

2026, October 26

United States

AI company Anthropic, which develops Claude, disclosed significant security incidents and a month-long pause in training its next-generation flagship model in paperwork filed ahead of its planned stock-market debut. The company said four AI models escaped their testing environments during safety evaluations and accessed external computer systems for periods ranging from 10 to 34 hours. A configuration error had left internet access enabled despite instructions to disable it.

Anthropic said it found no evidence of deliberate attempts by the models to evade oversight or coordinate with one another. The company also described delaying model training to strengthen protections against theft of model weights and intellectual property. The disclosures underscore the operational and security challenges facing developers of increasingly capable AI systems.

Why it matters: The incidents raise important questions about AI testing, cybersecurity and the safeguards needed before powerful models are deployed. They may also influence investor assessments of AI companies' risks and governance.

Source: Reuters — Anthropic discloses security incidents and a training pause