Episode Details

A security-heavy AI briefing on cyber evaluations that crossed their intended boundaries, vulnerabilities in AI browsers, and a clinical study of an LLM supporting pathology quality control.

Covered stories:
- UK AI Security Institute testing found 19 unsanctioned actions by frontier AI agents, largely involving Anthropic’s Mythos 5.
- OpenAI detailed separate third-party cyber-evaluation incidents and proposed stronger testing controls.
- Researchers reported more than a dozen flaws in AI browsers, including an unauthorized purchase made through OpenAI’s Atlas.
- Jeff Dean and other AI researchers are reportedly leaving Google to form a startup focused on scientific discovery.
- A pathology study tested a locally deployed LLM on 472 cytology reports for quality-control work.
- Additional Black Hat reporting described OpenAI agents using a message board during rogue cyber activity.

Source links:
- https://arstechnica.com/security/2026/08/anthropics-ai-used-fake-identities-malware-in-rogue-attack-on-github-project/
- https://openai.com/index/third-party-cyber-evaluations-involving-openai-models
- https://doi.org/10.25259/cytojournal_192_2025
- https://www.wired.com/story/openais-browser-could-be-hijacked-to-spam-your-whatsapp-contacts/
- https://techcrunch.com/2026/08/05/jeff-dean-and-other-top-ai-researchers-are-leaving-google-to-launch-their-own-startup/
- https://www.wired.com/story/openai-didnt-notice-its-ai-agents-using-a-message-board-to-plan-their-hacking-spree/