Threat Actors Tried Using Claude AI To Make Bioweapons, Anthropic Reveals In New Report

Anthropic on Thursday said it has disrupted multiple attempts by scientists to use its advanced artificial intelligence models to assist in the potential development of biological weapons, highlighting a terrifying new frontier in global security threats.

The alarming disclosure from the company's threat intelligence report arrives amid a high-profile internal crisis. On Sept. 9, Anthropic researcher Jacob Coxon publicly resigned in protest, stating that major AI labs are "racing straight to self-improving superintelligence and gambling with our lives". Coxon warned that those building these systems earnestly believe the technology could lead to human extinction by the end of the decade.

The Gain-of-Function Threat

The most severe biological threat detailed by Anthropic occurred in May. A scientist attempted to use the Claude AI model to draft a grant application for gain-of-function research on the chikungunya virus, a mosquito-borne disease known to cause debilitating joint pain.

The user sought to engineer mutations that would make the virus significantly more harmful by repeatedly infecting live animals. While such techniques are sometimes used legitimately to develop vaccines, Anthropic flagged the activity as highly suspicious because the research was intended for a military institute.

The company permanently banned the users involved after discovering they had actively tried to bypass geographic restrictions and hide their true intentions.

ALSO READ: '10% Doesn't Seem Unreasonable': Geoffrey Hinton On Prediction AI Could Kill All Humans Within A Decade

An "Incredibly Nuanced" Danger

Monitoring AI for biological warfare is exceptionally difficult because the line between legitimate medical breakthroughs and malicious pathogen engineering is razor-thin.

Jacob Klein, Anthropic's head of threat intelligence, told The New York Times that the threat is rarely as obvious as a comic-book villain declaring a desire to wipe out humanity. "It's an incredibly nuanced situation," Klein said, adding that a military institution conducting gain-of-function research inherently raises red flags.

Internal Crisis over 'Extinction' Risks

The bioweapons revelations lend immediate, real-world weight to the existential fears gripping Silicon Valley.

Coxon's abrupt departure this week sent shockwaves through the industry, particularly after his claims were corroborated by senior staff. Evan Hubinger, Anthropic's Alignment Science Lead, publicly backed Coxon's assessment on X, stating that he personally believes there is a greater than 10% chance that AI will kill all humans within the next decade.

The growing internal dissent suggests that even the architects of frontier AI models fear the technology is advancing faster than their ability to implement adequate safety guardrails.

Broader State-Sponsored Threats

Beyond the biological realm, Anthropic's threat report cataloged a litany of state-sponsored misuse.

The company thwarted attempts by actors in China, Russia, and Yemen to use Claude for conventional weapons development, including the design of missiles, armed drones, and bombs. In one notable case, an Iran-backed Houthi militia in Yemen reportedly consulted the AI to troubleshoot a guided rocket launch failure.

Security experts are increasingly demanding that frontier AI models be restricted from adversarial nations. Andrew Weber, a former assistant secretary of defense, reportedly called the report's findings "chilling." He stressed that because nations like Russia, China, and North Korea actively pursue prohibited biological weapons, it is imperative that their researchers are denied access to these highly capable AI systems.

ALSO READ: 'Gambling With Our Lives': Anthropic Researcher Quits, Says 'AI Could Kill Us All By End Of Decade'

Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.