Anthropic blocks AI attempts to engineer biological and conventional weapons
A new threat report reveals that advanced AI models now possess the scientific capabilities to assist in developing pathogens and weapon software.
Anthropic has disrupted multiple attempts to use its AI models to facilitate the development of biological and conventional weapons. In a threat intelligence report released in September 2026, the company detailed these interventions occurring over an eight-month period between December 2025 and August 2026.
According to the report, Anthropic blocked a specific attempt in May involving a researcher affiliated with a military institute. The individual sought to use Claude AI to draft a grant proposal aimed at engineering mutations of the chikungunya virus to increase its harmfulness in live animals. Beyond biological threats, the company identified a new category of abuse: attempts to use its technology to develop software for conventional weaponry, including bombs, missiles, firearms, and armed drones. The report detailed six such cases, with three occurring in China, two in Russia, and one in Yemen.
The Capability Gap
This shift in threat profiles is driven by the rapid evolution of large language models. Anthropic noted that while its older models were not meaningfully helpful for dangerous biological research, newer iterations possess the advanced scientific capabilities required to assist in complex and potentially hazardous work. This transition marks a critical turning point in AI safety, as tools once deemed too limited for such misuse have now crossed a functional threshold.
Security Implications
The findings highlight a growing security gap created by the "AI race," where the barrier to creating weapons of mass destruction may be lowered by accessible high-compute models. This creates a profound "dual-use" dilemma: the same AI capabilities that accelerate vaccine development and legitimate medical breakthroughs can be repurposed for pathogen engineering. Consequently, the responsibility for policing global security and preventing the proliferation of biological threats is increasingly falling upon private AI corporations rather than traditional state regulatory bodies.
Future Outlook
As AI models continue to gain scientific proficiency, the industry faces mounting pressure to implement more robust safeguards. The report arrives amid heightened anxiety over the existential risks of AI, coinciding with the resignation of an Anthropic researcher who warned that the current trajectory of the AI race could lead to human extinction. While Anthropic has successfully disrupted these specific bids, the company's findings suggest that the potential for misuse is expanding alongside the models' intelligence.