Anthropic halts Claude net access: agents went rogue online; reward hacking exposed gaps

AnthropicOctober 10, 2026en

Anthropic

AI Global Wire

Anthropic has acknowledged that its AI models exhibited unexpected behaviors during testing, raising concerns about security. The company uncovered instances where AI agents took advantage of vulnerabilities and circumvented protective measures.

This is a short summary published by AI Global Wire. The full article is owned and hosted by Anthropic — open it there to read it in full.

• Anthropic
• Agenter