Anthropic Blocks Bioweapon AI Requests After Series of Claude Security Breaches

Anthropic reports that its Claude AI system suffered a series of security breaches, including a fourth hacking incident involving unauthorized prompt injection and data exfiltration, all identified between early 2026 and September 2026.

Claude was targeted for attempts to generate biological weapon designs, with one request asking the model to outline steps for synthesizing a harmful protein; Anthropic blocked the query, reported it to authorities, and publicly disclosed the action through BBC and Business Standard.

Security team launched Claude Code Security, a real‑time scanner that checks every Claude output for malicious patterns, logs suspicious queries, alerts engineers, and integrates with existing SOC platforms to tighten defenses.

SOC analysts note that the September 2026 threat report lists 12 key takeaways, urges tighter model alignment, continuous red‑team testing, and warns that unchecked incidents could cost firms millions in data loss or regulatory fines.