Anthropic has identified and blocked multiple attempts by malicious actors to use its Claude AI models for dangerous purposes, including biological weapons research, cyberattacks, and surveillance operations. The company published its third misuse report since March 2025, detailing cases ranging from state-sponsored groups to lone operators attempting to exploit AI capabilities for harm. The findings underscore a critical challenge facing the AI industry: as models become more powerful, they become more attractive targets for misuse, and older safeguards are no longer sufficient.

What Types of Misuse Did Anthropic Actually Stop?

Between December 2025 and August 2026, Anthropic researchers identified misuse attempts across several categories. The company documented cases involving actors ranging from spyware vendors and politically motivated individuals to state-sponsored groups. One particularly concerning case involved an unnamed actor attempting to use Claude to help write a grant application for gain-of-function research on the chikungunya virus, a mosquito-borne pathogen that causes severe pain and fever.

The grant proposal sought to enhance the virus's transmissibility and ability to evade immune responses. While such research could theoretically support vaccine development, Anthropic noted it could equally be used to make the pathogen more dangerous. The company's systems blocked this request before it could proceed. Beyond biological threats, Anthropic also identified:

• Cyberattack Planning: Actors attempting to use Claude to develop sophisticated attacks that would have previously required specialized technical expertise or state-level resources.
• Surveillance Operations: Groups seeking to leverage AI for monitoring and tracking activities that could harm individuals or organizations.
• Influence Campaigns: Nine documented cases originating in Russia, Iran, Turkey, and across the Persian Gulf, South Asia, Africa, and Europe, where actors created hundreds of fake social media accounts to amplify political messaging over the course of a week.
• Model Distillation: An industrial-scale, covert campaign to extract Claude's capabilities and replicate them in another model without authorization.

Why Are Older Claude Models Less Protected Than Newer Ones?

A key insight from Anthropic's report reveals a generational gap in AI safety. The company's older models, such as Claude Opus 4 and Claude Sonnet 4.5 from 2025, were considered too limited to meaningfully assist sophisticated users in conducting dangerous biological research. As a result, safeguards on these earlier versions were less stringent, focused primarily on preventing access to content that might help novices recreate known bioweapons.

However, Anthropic's newer Claude Fable and Mythos-class models represent a significant leap in capability. These advanced models can assist with complex scientific research tasks that older versions could not handle. This shift forced Anthropic to reconsider its safety assumptions. The company stated: "The evidence is no longer certain, and we cannot make that same assurance" that newer models would be unable to assist in dangerous biological research.

Notably, none of the misuse cases Anthropic documented in its report involved the newer Claude Fable or Mythos-class models, with one exception: the industrial-scale model distillation campaign. This suggests that either the newer models' safeguards are working, or that malicious actors have not yet discovered how to exploit them at scale.

How Is Anthropic Strengthening Its Defenses?

In response to these findings, Anthropic has implemented stronger safeguards in its more recent models, particularly Claude Fable 5. The company has applied restrictions that limit access to a wide range of dual-use biological research queries, meaning questions that could serve legitimate scientific purposes but also pose risks if misused.

Beyond technical safeguards, Anthropic has taken a transparency-first approach. The company published detailed findings from its misuse report, including snippets of malicious code and AI prompts it discovered, to help other AI developers and governments identify and prevent similar abuse. Anthropic stated: "We're publishing this work because we believe we have a responsibility to disclose malicious misuse of our services. As models become increasingly capable, their risks will increase, unless AI developers and society's defenders act to make them safer".

Anthropic

The company also shared information with government authorities and industry partners to strengthen collective defenses. This collaborative approach reflects a growing recognition that no single company can address AI safety alone.

What Do Experts Say About AI Safety Responsibility?

The release of Anthropic's misuse report comes amid broader concerns about AI safety and corporate responsibility. John Thickstun, an assistant professor of computer science at Cornell University, highlighted a fundamental tension in the current AI governance landscape.

"It is an uncomfortable position for companies like Anthropic and OpenAI to be in when they are expected to determine what is safe vs. unsafe behavior and make value judgments at societal scale without any kind of democratic or deliberative oversight," Thickstun stated.

John Thickstun, Assistant Professor of Computer Science, Cornell University

This concern has taken on added urgency following the resignation of Jacob Coxon, an Anthropic researcher who announced his departure amid fears that Anthropic and its rival OpenAI "are racing straight to self-improving superintelligence and gambling with our lives." Coxon warned that some colleagues now believe AI could threaten human life by the end of the decade, raising questions about whether industry self-regulation is sufficient.

Why Does This Matter for the Future of AI Development?

Anthropic's findings illustrate a paradox at the heart of modern AI development. As models become more capable and useful for legitimate purposes, they simultaneously become more useful for harmful ones. The company noted that elaborate cyberattacks no longer require sophisticated skills; even lone individuals can now create threats that would have been impossible just a year ago.

This democratization of AI-powered harm has significant implications for how companies, governments, and society approach AI safety. It suggests that traditional approaches to security, which often rely on obscurity or technical barriers, may be insufficient. Instead, a multi-layered approach combining technical safeguards, transparency, information sharing, and potentially new regulatory frameworks may be necessary.

Anthropic is planning an initial public offering this fall, which will likely intensify scrutiny of its safety practices and governance. The company's proactive approach to documenting and addressing misuse may serve as a model for how AI developers can balance innovation with responsibility, though experts like Thickstun argue that industry-led oversight ultimately requires democratic input and government regulation to be truly effective.