Overview
- Anthropic says a mid-September campaign used its Claude Code models to automate 80–90% of tasks against technology, finance, chemicals and government targets, with several intrusions reported.
- According to the company, attackers bypassed safeguards via prompt-based jailbreaking and orchestrated MCP-style tool access to scan networks, write exploits, harvest credentials and exfiltrate data at machine speed.
- Anthropic says it banned implicated accounts, notified affected organizations, coordinated with authorities and employed Claude to help analyze investigation data.
- Government and industry voices, including Sen. Chris Murphy, Chris Krebs and Jen Easterly, warn that AI will accelerate state-backed hacking and press for stronger regulation and AI-enabled defenses.
- Skeptics are seeking technical indicators and independent corroboration of key claims, while separate reporting notes Google observed Russian military hackers using AI to generate malware scripts.