AI Safety Report Reveals Blocked Bioweapon and Espionage Threats
Devangi Chandok, Jadetimes Staff
In a alarming reveal of how artificial intelligence is reshaping national security, frontier AI developer Anthropic announced it has disrupted multiple attempts by bad actors to leverage its flagship chatbot, Claude, to assist in developing biological weapons.
The revelations, detailed in the California-based startup’s 154-page threat intelligence report, mark the first time a major AI firm has publicly confirmed that real-world actors attempted to use its systems for potential biological warfare research.
A Dual-Use Dilemma
Between December 2025 and August 2026, Anthropic identified five specific cases where users tried to deploy Claude for research that could support biological weapons development.
In one instance, a user attempted to use the AI to draft a grant application for gain-of-function research on the chikungunya virus, a mosquito-borne pathogen. The proposed work involved engineering viral mutations designed to increase harmfulness in live animals, allegedly intended for a foreign military research institute. In another case, a researcher used the AI to draft work on orthopoxviruses, the virus family that includes smallpox.
"Biological misuse is one of the most serious risks of frontier AI models," Anthropic noted in its report, warning that without stringent safeguards, such capabilities "could have catastrophic consequences."
However, identifying clear intent remains a major roadblock for safety teams. Because the underlying science for cure development often mirrors the science used to engineer pathogens, determining whether a query is dangerous presents a massive challenge.
"You are not seeing someone in a comic book kind of way say, 'Hey, I want to build a biological weapon to kill everybody,'" Jacob Klein, Anthropic’s head of threat intelligence, said in an interview. "It's an incredibly nuanced situation."
Beyond Bioweapons: Cyber Espionage and Warfare
The biological threats were part of a broader, troubling array of misuse detailed in the report. Anthropic disclosed that foreign state-backed actors and cybercriminals have aggressively targeted its systems across seven harm areas, including cyber operations, influence campaigns, mass surveillance, and conventional weapons design.
Cyberattacks: A hacking group consistent with Russia-linked "Midnight Blizzard" used Claude to build a self-correcting malware tool. The system automatically detected when security defenses flagged its code and rewrote the software until it evaded detection.
Conventional Arms: Anthropic documented six cases where Claude was used to write software for firearms, missiles, armed drones, and targeting systems. In one case, a non-state actor in Yemen test-fired a guided rocket and consulted Claude hours later to analyze why the launch failed.
State Surveillance: Iranian and Chinese state-aligned actors used the models to track dissidents and score social media posts for political sensitivity. In Mali, a single consultant used Claude to engineer a surveillance system monitoring 25 million SIM cards.
Google reported a similar incident, revealing that a user attempted to extract a "complete, step-by-step technical guide for synthesizing weaponized biological agents" from its Gemini AI tool.
Political Friction and Demands to Pause
The sudden wave of disclosures has intensified the political battle over AI governance.
In Washington, U.S. Senator Bernie Sanders pushed back against the rapid pace of development, introducing legislation to pause advanced AI research and place a ban on artificial superintelligence. "When scientists tell you there is a chance that it could have a cataclysmic impact on humanity, you've got to be a moron not to say, 'slow it down,'" Sanders stated.
Conversely, President Donald Trump rejected calls for a shutdown, emphasizing national competitiveness. "If we don't win AI, we're going to be put in a very bad position," Trump said, dismissing extinction fears.
With companies continually updating their guardrails to patch vulnerabilities, the report serves as a stark reminder: as AI grows more capable, the race between defensive safety filters and malicious intent is rapidly tightening.
Comments