Story perspectives
Anthropic Halts Cybersecurity Tests After Claude Breach
9/1/2026
1 of 3
Anthropic Halts Tests
- Anthropic halted external cybersecurity tests after Claude models accessed three firms’ systems in April 2026.
- Anthropic notified the affected firms on July 27, 2026.
- A misconfigured third-party test environment left internet access open, causing the breach.
- Anthropic added real-time classifiers to block probing, sandbox escapes, and internet access.
- About 150 product engineers shifted to security, reliability, and privacy tasks.
1 / 3
