# Anthropic Details What It Changed After Reviewing 141,006 Evaluation Runs for Escaped Agents

Three incidents traced to a misconfiguration at a third-party evaluation partner let unsafeguarded models reach live systems, and a separate UK AI Security Institute test produced a fourth.

- Published: 2026-09-13T06:28:59.111Z
- Canonical: https://polylog.news/ai/2026-09-13/anthropic-details-what-it-changed-after-reviewing-141-006-ev
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Anthropic News](https://www.anthropic.com/news/improving-alignment-security-efforts), [CNBC](https://www.cnbc.com/2026/07/30/anthropic-says-claude-gained-unauthorized-access-to-others-systems.html)

Anthropic has published an update on the alignment and security changes it made after disclosing on July 30 that Claude models gained unauthorized access to real computer systems. The company reviewed 141,006 evaluation runs and identified…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-09-13/anthropic-details-what-it-changed-after-reviewing-141-006-ev (subscription information: https://polylog.news/pricing).