anthropic-claude-code-abuse-2025 · Disclosed 2025-11-13
Claude Code: attacker misuse in a multi-organization intrusion campaign
Anthropic reported attacker use of Claude Code for reconnaissance, exploit development, credential harvesting, and exfiltration. About 30 organizations were targeted, with success reported at a small number.
Outcome: Confirmed breach
Entry path and evidence
- Reported fact
Anthropic reports investigating attacker misuse of Claude Code in an observed campaign, distinct from evaluation incidents.
[s1]Introduction / How the cyberattack worked - Reported fact
Human operators selected targets and disguised fragmented tasks as defensive testing; Anthropic describes AI assistance through reconnaissance and exfiltration.
[s1]How the cyberattack worked
Timeline
Incident disclosed. [s1]
Reported response
- Reported fact
Anthropic reported banning accounts, notifying affected organizations, coordinating with authorities, and improving safeguards.
[s1]Introduction
Evidence relevant to prevention
Insufficient evidence
Victim-specific entry causes are unspecified; inspect credential reach, access logs, and permissions of AI operated in your own environment.
Editorial assessment; not a determination of liability. [s1]
Unknowns and AI involvement
Anthropic reports attacker use of Claude Code; this does not establish unavoidable victim vulnerabilities or a population-wide increase in AI attacks.
Victim names, vulnerabilities, CVEs, and precise entry dates are undisclosed; target count is not victim count, and autonomy assessments are the investigator’s interpretation.
Sources
[s1] Anthropic · Primary source
Disrupting an AI-orchestrated cyber espionage campaign ↗Published 2025-11-13 · Reviewed 2026-10-02