Lauge Risom Koch 
Anthropic published a security report detailing how it blocked malicious actors from using its AI models for biological weapons research and coordinated disinformation.
Anthropic announced it successfully blocked malicious actors from utilizing its artificial intelligence models for dangerous activities. According to AP News, the blocked efforts involved potential cyberattacks, illicit surveillance, and research linked to biological weapons. The technology company published these findings to highlight novel threat activities identified between December 2025 and August 2026.
The recent safety report details specific attempts to weaponize scientific research using the AI platform. As reported by AP News, unknown actors requested assistance from the Claude system to author a grant application involving the chikungunya virus. The proposed gain-of-function research aimed to genetically alter the mosquito-borne pathogen to enhance its transmissibility and immune evasion properties.
The company noted that while such research can develop treatments, it also carries severe security risks. According to AP News, Anthropic stated that this specific research “could also be used to make the pathogen more dangerous.” Developers subsequently implemented stronger safeguards restricting access to dual-use biological queries in newer models like Claude Fable 5.
Exposing coordinated disinformation operations
The security report also exposed multiple state-sponsored groups attempting to generate coordinated propaganda networks. According to AP News, investigators found threat actors creating hundreds of fake social media profiles designed to resemble ordinary citizens. These automated accounts systematically amplified specific political viewpoints across various platforms over short timeframes.
Anthropic identified nine separate disinformation operations originating from regions including Russia, Iran, and South Asia. As noted by AP News, the company emphasized its unique ability to detect these influence operations during their initial construction phase. This early detection capability allows developers to intervene before the generated propaganda actively circulates on social networks.
The technology firm shared these behavioral patterns with government authorities to strengthen collective defenses. According to AP News, the company urged competitors to identify and prevent similar malicious misuse within their own systems. The report explicitly included snippets of malicious code and AI prompts to assist other developers in recognizing these threats.
Facing internal and external scrutiny
The publication of this threat report coincides with growing internal apprehension regarding rapid technological advancement. According to AP News, Anthropic researcher Jacob Coxon recently resigned over fears regarding the industry’s trajectory. He publicly warned that major developers are “racing straight to self-improving superintelligence” while gambling with human lives.
Coxon highlighted severe existential concerns shared among several colleagues working within the artificial intelligence sector. As reported by AP News, his resignation post noted that some employees believe AI could directly threaten human life by the end of the decade. These internal warnings echo broader industry fears regarding the technology’s potential to eventually elude human control.
External experts continue to question whether private companies should unilaterally govern these increasingly powerful systems. According to AP News, Cornell University professor John Thickstun argued that companies lack the democratic oversight required to make societal value judgments. Academics advocate for formal government regulation to manage these risks rather than relying on the technology industry to police itself.
