Anthropic Blocks Attempts to Misuse AI: “Biochemical Weapons Research and Cyberattack Defense”

[Financial News] Anthropic, an artificial intelligence (AI) startup, said on the 11th local time that it had blocked attempts to use its AI models for malicious purposes, including cyberattacks, surveillance, and research with potential applications in biochemical weapons development.
U.S. media reported that Anthropic disclosed in its third report on AI misuse since March 2025 that it had blocked a request to draft a grant application for research related to gain-of-function studies aimed at enhancing the Chikungunya virus’s transmissibility and ability to evade the immune system.
As AI models become dramatically more capable, concerns are growing that the threat landscape is changing to the point where a single individual can carry out cyberattacks or other dangerous activities that once required sophisticated technical expertise.
Chikungunya is a mosquito-borne virus that causes severe pain and high fever. Anthropic explained, "Although the research could be used to develop vaccines, it also carries a high risk of being misused as a biochemical weapon to make the pathogen more dangerous."
Anthropic also detected and blocked nine cases in advance in which state-backed actors from countries including Russia, Iran, and Türkiye allegedly used generative AI to create hundreds of fake social media accounts and spread specific political views.
Anthropic has applied enhanced safeguards to its latest models, including the recently released Fable 5, to restrict queries related to dual-use biological research with military and civilian applications. This is because the risks increase sharply as newer AI systems with greater intelligence surpass earlier models such as Claude Opus 4 and Claude Sonnet 4.5.
The report was released two days after Jacob Coxon, a researcher at Anthropic, submitted his resignation ahead of the company’s initial public offering (IPO) this fall. In his resignation letter, Coxon warned, "Anthropic and its competitor OpenAI are focusing solely on competing to develop superintelligent AI without ensuring safety."
Anthropic said, "The more powerful AI models become, the greater the risks become as well," adding, "We will share information about misuse cases with governments and other technology companies to strengthen a joint defense system."

[email protected] Jae-jun Yoon Reporter