Anthropic Blocks AI Biological Weapons Misuse
Anthropic has disclosed that it halted attempts to use its artificial intelligence systems for biological research, cyberattacks, and coordinated influence campaigns.
Artificial intelligence developer Anthropic has blocked attempts by malicious actors seeking to deploy its models for cyberattacks, surveillance, and dual-use research that could assist in developing biological weapons. In its third report detailing AI misuse since March 2025, the company stated that expanding model capabilities have steadily reduced the technical barriers required for digital threats, allowing even lone individuals to generate attacks that were not feasible a year prior.
Between December 2025 and August 2026, researchers at the firm uncovered unauthorized activity carried out by spyware vendors, politically motivated individuals, and state-backed groups. Among the specific incidents documented was a blocked attempt to use the Claude model to author a scientific funding grant. The proposal involved gain-of-function research on the mosquito-borne chikungunya virus aimed at enhancing its transmissibility and immune evasion properties, which could potentially be used to make the pathogen more dangerous.
"As models become increasingly capable, their risks will increase, unless AI developers and society’s defenders act to make them safer."
Anthropic, which is planning an initial public offering this fall, explained that older 2025 models such as Claude Opus 4 and Claude Sonnet 4.5 featured less stringent safeguards because they sat below the threshold of being able to meaningfully assist with dangerous biological tasks. In response to advancing scientific capabilities, the company implemented tighter restrictions on dual-use biological queries in its newer models, including Claude Fable 5. Anthropic also identified nine covert operations from regions including Russia, Iran, Turkey, Europe, and South Asia that attempted to create hundreds of fake social media accounts to amplify political views.
The disclosure comes amid heightened scrutiny over industry oversight and safety controls. The report followed the resignation of Anthropic researcher Jacob Coxon, who expressed concern that leading artificial intelligence developers are moving too quickly toward superintelligence. Outside experts, including Cornell University computer science assistant professor John Thickstun, noted that tech firms currently face the uncomfortable task of determining safe versus unsafe behavior at a societal scale without democratic or deliberative oversight.
Important notes:
The report details malicious activity identified by Anthropic between December 2025 and August 2026.
None of the misuse incidents involved Anthropic's newer Claude Fable or Mythos-class models, except for a single unauthorized campaign aimed at covertly replicating model capabilities.
Anthropic shared malicious code snippets and intercepted prompts to help government authorities and industry competitors strengthen collective security defenses.