Anthropic, the AI company behind the Claude model, has confirmed that it identified and stopped multiple attempts to use its technology in the creation of biological weapons.
The announcement came in its latest threat‑intelligence report, published on Thursday, and is part of a growing wave of warnings from AI firms about the potential dual‑use threats of advanced language models.
According to the report, between December 2025 and August 2026, the company disrupted the use of its Claude Haiku, Sonnet, and Opus models in activities ranging from fake dating apps and hotel Wi‑Fi scams to surveillance aimed at identifying political dissidents. Importantly, “five case studies” detailed actors who sought to use Claude to facilitate the synthesis of weaponised biological agents – a reminder that the same technology that can write a scientific paper can also compose a dangerous protocol.
Many of the misuse incidents involved state‑backed actors. The report names a Russia‑based cyber‑espionage campaign and an Iranian propaganda institution as well as Chinese labs attempting to replicate Claude’s capabilities. One group, “ShinyHunters,” was flagged for using the model to develop an automated system that rewrote malware to evade security defenses.
Anthropic’s head of threat intelligence, Jacob Klein, told the New York Times that “the situation is incredibly nuanced.” He emphasised that users are not “building a biological weapon to kill everyone” in a comic‑book sense, but that the information being generated could double‑serve weapon and cure development.
“When people can produce detailed step‑by‑step protocols for synthesising biological agents,” Klein said, “the same knowledge can help produce vaccines or treatments.” The report argues that without robust safeguards, large‑scale AI models could have catastrophic consequences.
These findings come on the heels of stronger calls for a pause on advanced AI development. U.S. Senator Bernie Sanders has introduced legislation to ban artificial super‑intelligence and temporarily halt cutting‑edge AI research. Meanwhile, President Donald Trump has dismissed fears, warning that “if we don’t win AI, we’ll be in a very bad position.”
Industrial peers are echoing Anthropic’s vigilance. Google recently warned of an opponent who requested a step‑by‑step guide to synthesise weaponised biological agents using its Gemini model. Anthropic’s report is part of a new norm where AI firms publish threat intelligence bulletins detailing how they identify and disrupt misuse.
Anthropic has also worked with law enforcement and industry partners, sharing threat intelligence where appropriate. The firm stresses that its work is only one facet of a larger global effort to ensure AI is used safely.
