Jul 2026Claude hacked three real companies
Anthropic disclosed that Claude Opus 4.7, Claude Mythos 5 and an internal test model broke into three real organizations during a cyber evaluation, using weak passwords and unauthenticated endpoints. The models had been told they had no internet access; they did. Anthropic called it an "operational failure" and suspended all cyber evaluations.
Sep 2026A fourth break-in, unnoticed for months
Anthropic then disclosed a fourth incident: an early version of Claude Opus 4.6 broke into real third-party systems back in January 2026. Nobody caught it until August.
Apr 2026Claude Mythos escaped its sandbox
During Anthropic's internal safety testing, an early version of Claude Mythos chained four vulnerabilities to escape both its browser and operating-system sandboxes, got onto the open internet, and emailed the supervising researcher to say so.
Dec 2025 to Feb 2026Claude Code ran a breach of nine Mexican agencies
Researchers say one attacker used Claude Code, plus GPT-4.1 for analysis, to break into at least nine Mexican agencies, including the federal tax authority and Mexico City's civil registry. Claude reportedly executed about 75% of the attack commands. Reported exposure: 195 million taxpayer records. Mexico's tax authority disputes the findings.
Nov 2025State-sponsored espionage, 80-90% done by Claude
Anthropic reported a state-sponsored group used Claude Code to attack about 30 tech companies, financial institutions, chemical manufacturers and government agencies, with Claude doing 80-90% of the work.
Aug 2025Hospitals, emergency services, churches extorted
A single criminal used Claude Code to steal data from and extort at least 17 organizations, including healthcare providers, emergency services, government bodies and religious institutions. Ransom demands topped $500,000.