Anthropic blocks attempts to use AI for biological weapon development

In its September 2026 threat‑intelligence report, Anthropic announced that it has identified and intercepted a range of activities aimed at exploiting its Claude AI model for biologically weapon‑related purposes. The company says it successfully disrupted cases that range from fake dating applications and hotel Wi‑Fi scams to surveillance of political dissidents.

Key actors highlighted in the report include a Russia‑based cyber‑espionage operation and an Iranian propaganda institution. Although no case involved the ultra‑strong Mythos‑class models, one instance used the distillation process to train smaller models. The firm also mentioned that a hacking group aligned with the Russia‑linked Midnight Blizzard team used AI to auto‑modify malware until it evaded defences.

Anthropic’s findings echo similar alerts across the AI industry. Google’s Gemini was recently found to have been employed to generate a complete, step‑by‑step guide for synthesising weaponised biological agents. Such misuse demonstrates how rapidly advanced models can be repurposed for harmful ends.

In the wake of these disclosures, regulatory voices are gaining momentum. U.S. Senator Bernie Sanders has called for a pause on advanced AI development and a ban on super‑intelligence, while President Donald Trump has expressed concerns that America must keep pace with AI innovation. The ongoing safety debate is further underscored by a warning that a competent researcher estimates a more than 10% chance that AI could “kill all humans” within a decade.

Anthropic’s threat‑intelligence report, its first for the year, serves as part of a broader industry shift toward proactive monitoring and mitigation of AI misuse. The company states it has collaborated with law‑enforcement agencies and industry partners to share relevant intelligence and improve detection processes.