Summary
In November 2025, Anthropic disclosed that it had detected and disrupted what it assessed to be the first documented large-scale cyberattack executed largely without human intervention. A threat actor — assessed with high confidence to be a Chinese state-sponsored group — jailbroke and manipulated an AI coding agent into reconnaissance, vulnerability discovery, exploit writing, credential harvesting, and data exfiltration against roughly thirty global targets. The AI performed an estimated 80–90% of the work, with humans intervening only at a handful of critical decision points.
Why it matters
The campaign turned on three converging capabilities: model intelligence, agency (running autonomously in loops), and tool access (often via the Model Context Protocol). It is the adversarial counterpoint to the agent-readable web standards, which assume well-behaved machine readers honoring advisory rules — this report shows what happens when agents act in bad faith at machine speed. That tension sharpens the Independent Internet’s emphasis on being safe by construction: airgap-first nodes, on-premises data, and post-quantum security, rather than trust in good behavior. This is a copyrighted report; only an original summary and the canonical link are kept here.
Part of the AP0110.ORG source library. See the wiki overview.
Full source text: read the original document.