A comprehensive 154-page threat intelligence report released by Anthropic on Thursday reveals that various malicious actors, including state-sponsored groups, criminals, and propagandists, have attempted to exploit its artificial intelligence models. These efforts range from designing conventional weapons and missiles to developing dangerous pathogens and conducting surveillance on dissidents.
In its findings, the company stated, “The cases we share here aren’t typical misuse, but rather examples of the most notable and novel threat activity we’ve identified to date.” Anthropic emphasized that it is disclosing these incidents because it holds a responsibility to address the malicious exploitation of its services.
The report highlights five specific case studies involving scientists who used AI models for biological research. In these instances, users reportedly bypassed safeguards designed to block access from “unsupported regions” and actively concealed the nature of their research. One notable case involved a scientist utilizing the Claude model to draft a state-sponsored grant application for studying the chikungunya virus, a move that raised concerns because the work was intended for a military research institute.
“Biological misuse is one of the most serious risks of frontier AI models,” the company noted, warning that without robust safeguards, such capabilities could lead to catastrophic outcomes. While the company has banned the identified accounts, it did not disclose the specific identities of the institutions involved, citing uncertainty regarding the researchers’ ultimate intent.
Beyond biological threats, the report documents a wide array of malicious activities. These include Russian espionage and cyber-attacks, surveillance programs targeting Uyghurs in Syria, and propaganda campaigns spanning Iran, Malaysia, Russia, and Bangladesh. Furthermore, the company identified the use of its Claude model in Russia, China, and Yemen to assist in the development of software for firearms, missiles, armed drones, and other munitions.
The publication of this data follows the resignation of former employee Jacob Coxon, who departed the company two days earlier. Coxon alleged that Anthropic and OpenAI are prioritizing a “race to self-improving superintelligence” that could lead to human extinction by 2030. His claims have received public support from some current staff members.
Industry experts, such as Heidy Khlaaf, chief AI scientist at the AI Now Institute, argue that immediate, real-world risks are more pressing than hypothetical future scenarios. “AI accelerationism and AI doomerism are two sides of the same coin: they both enforce the notion that an AGI superbeing will come into existence,” Khlaaf remarked, adding that technology capable of facilitating cyber warfare and weapons production poses a more urgent danger.
Anthropic maintains that these risks are often investigated internally or by governments and academics rather than being visible to the public. The company concluded that the entire AI sector must collaborate with global authorities to establish stronger defenses. “As models become increasingly capable, their risks will increase, unless AI developers and society’s defenders act to make them safer,” the report stated. The report also notes that anthropic told the New York Times this case was especially concerning because it could see the study was to be done at a military research institute. The report also notes that similar to dengue and malaria, that can cause months of severe pain, fever and other debilitating symptoms, the virus is a mosquito-borne disease. The report also notes that current Anthropic employees posted public agreements with him. The report also notes that like those detailed in Anthropic’s intelligence report, are far more concerning within the next three years than an apocalypse brought on by an omnipotent intelligence, many AI experts say the real-world threats.
Comments