Anthropic has lifted the lid on a disturbing episode in which unknown actors tried to co‑opt its cutting‑edge Claude model to draft a grant proposal aimed at genetically super‑charging the chikungunya virus. The incident, uncovered between December 2025 and August 2026, marks one of the most alarming examples of AI misuse the firm has documented since its first safety report in March 2025. The revelation comes as the AI race intensifies and as the company prepares for a blockbuster IPO this autumn.
A Close Look at the Threat
The suspicious request was blocked by Claude’s latest safeguards, but the attempt itself is a stark reminder of how powerful AI tools can be weaponised. In the grant draft, the chikungunya virus—an aggressive mosquito‑borne pathogen causing severe fever and debilitating joint pain—was earmarked for “gain‑of‑function” research. The aim, according to the proposal, was to enhance the virus’s transmissibility and its ability to dodge immune responses, effectively making it more dangerous. While such studies can, in theory, inform vaccine development, the same techniques could also be used to craft a potent biological weapon.
Anthropic’s report highlights that the request never reached its intended audience. The firm’s advanced models—Claude Opus 4, Claude Sonnet 4.5, and the newer Claude Fable 5—are equipped with layered filters designed to stop dangerous biological research queries before they can be acted upon. The older models, the company notes, were “well below the threshold where they could meaningfully assist a sophisticated user in carrying out dangerous biological research.” However, newer models capable of handling complex scientific tasks now sit at a more precarious edge, prompting Anthropic to tighten its defensive measures.
The incident is just one of nine cases the firm uncovered involving coordinated social‑media manipulation. Hundreds of fake accounts, masquerading as ordinary users, were used to amplify political narratives across Russia, Iran, Turkey, the Persian Gulf, South Asia, Africa, and Europe. Anthropic’s researchers caught these operations while they were still in the planning stages—a window of opportunity that traditional social‑media platforms often miss.
How Anthropic Is Fighting Back
Since the threat was identified, Anthropic has taken decisive action. Each malicious attempt was immediately blocked, and the company has reinforced its safety protocols, especially on the latest Claude Fable 5 model. “Because of this, Anthropic has applied stronger safeguards that restrict access to a wide range of dual‑use biological research queries,” the report states. These safeguards are not static; they evolve as the technology advances, reflecting the firm’s commitment to staying one step ahead of bad actors.
The safety enhancements are part of a broader strategy that includes sharing intelligence with government bodies and industry partners. By publishing detailed snippets of malicious code and prompts, Anthropic hopes to arm other developers with the knowledge needed to spot similar patterns on their own platforms. The firm’s third misuse report since March 2025 underscores a philosophy that transparency, when handled responsibly, can serve as a powerful deterrent.
The Wider Landscape of AI Misuse
Anthropic’s findings paint a picture of a rapidly shifting threat environment. From spyware vendors to “politically motivated individuals” and state‑sponsored propaganda machines, the actors exploiting AI are increasingly diverse. The firm’s researchers also uncovered an “industrial‑scale, covert campaign to extract a model’s capabilities and replicate them in another model without authorisation.” This illicit distillation case highlights how determined adversaries are willing to go to great lengths to harness AI’s potential for harmful ends.
The company’s stance is clear: as models become more capable, the associated risks grow. “We’re publishing this work because we believe we have a responsibility to disclose malicious misuse of our services,” Anthropic declares. “As models become increasingly capable, their risks will increase, unless AI developers and society’s defenders act to make them safer.” This call to action resonates beyond Anthropic, urging the entire AI ecosystem to prioritize safety alongside innovation.
Jacob Coxon’s Resignation Sparks Industry Alarm
The timing of the report coincides with a dramatic exit from the company. Jacob Coxon, a prominent AI researcher, announced his resignation amid fears that Anthropic and its chief rival, OpenAI, are “racing straight to self‑improving superintelligence and gambling with our lives.” In a candid post, Coxon warned that some colleagues now believe AI could threaten human existence by the end of the decade. His departure has ignited fresh debate about the ethical boundaries of AI development and the need for robust oversight mechanisms.
Anthropic’s response to Coxon’s concerns has been to double down on safety. The firm asserts that it has blocked each malicious activity identified, used those lessons to tighten safeguards, and shared information with authorities. “We hope that the findings in this report will help other developers recognise similar patterns on their own platforms, give governments and civil society a clearer view of how emerging threats take shape, and strengthen collective defences,” the company says.
Why it Matters
The episode underscores a critical truth: cutting‑edge AI tools are a double‑edged sword. While they empower developers to solve complex problems, they also provide a playground for those intent on causing harm. Anthropic’s proactive stance—blocking dangerous queries, tightening safeguards, and sharing intelligence—offers a blueprint for how AI companies can balance innovation with responsibility. As the technology race accelerates, the industry, regulators, and civil society must stay vigilant, ensuring that progress does not come at the cost of global safety. The stakes could not be higher, and the actions taken now will shape the future of AI’s role in society.
