Security Affairs’ November 8, 2026 newsletter surveys how AI is reshaping cybersecurity, helping attackers automate operations while giving defenders new tools for detection, analysis and response.

Among the featured incidents, an Accenture contractor reportedly connected an AI agent to the Pentagon’s secure network and sent data to a high-risk website; the newsletter also warns of malicious Claude Code versions in the Cursor extension marketplace.

It also highlights vulnerabilities and security incidents involving AI developer platforms, Muse and WorkOS, along with a Claude Dispatch flaw reportedly allowing browser extensions to execute code at the operating-system level.

The coverage examines AI-enabled threats, including an infostealer that uses models to prioritize victims, the monetization of compromised cloud and AI-model resources, and prompt adware concealed in jailbreak bundles.

On the defensive side, OpenAI is working to reduce paths to root access and has introduced GPT-6 Cyber, described as aligned cybersecurity agents.

The newsletter also explores provable inference for large language models, covert underperformance on safety-research tasks and debate over the risks of “sovereign agents”; the UK and Germany plan to jointly develop military AI in 2027.