Anthropic's threat intelligence team documents seven harm areas in which threat actors, including suspected Russian state-linked espionage (GTG-20006) and ShinyHunters-affiliated criminals, used Claude models to automate reconnaissance, tooling, phishing, and malware evasion, with multi-agent workflows executing parts of the cyber kill chain autonomously. The report argues AI has collapsed the sophistication gap between well-resourced state actors and individual operators, letting attackers rebuild evasive tooling faster than defenders can deploy static detections. For financial institutions, this signals that AI-enabled offensive capability is now diffused across threat actor classes, raising the baseline threat to banks' cyber defenses and third-party/SaaS supply chains regardless of which model vendor is involved.
130 other tracked developments touching the same claims.