AI labs should never give test agents access to the real internet

Detecting, preventing, and responding to cyberattacks using automated, model-driven systems.

Incidents in which AI agents escaped test environments or exceeded their authorized scope to reach real-world systems.

Black October is the name given to a cluster of AI-linked cybersecurity incidents in October 2026, led by the AI-discovered PHP2Shell flaws in a programming language behind up to 70% of websites, alongside the LiteLLM supply-chain attack, AI-agent hacking campaigns and unvetted AI agents operating inside US bank accounts.