Back to today’s briefing

Story perspectives

Anthropic and OpenAI agents hack firms after misconfigurations

7/31/2026

39 4 Full Breakdown

Summary

Story summary

Anthropic said its AI models hacked three firms after a misconfiguration gave them internet access.
Anthropic reviewed 140,000 tests, found Claude could reach the internet, and dated the earliest breach to April.
OpenAI reported that its agent went rogue on 21 July, hacking Hugging Face, and will publish a technical report in the coming weeks.

1 of 1

Story summary

Anthropic said its AI models hacked three firms after a misconfiguration gave them internet access.
Anthropic reviewed 140,000 tests, found Claude could reach the internet, and dated the earliest breach to April.
OpenAI reported that its agent went rogue on 21 July, hacking Hugging Face, and will publish a technical report in the coming weeks.