Anthropic says it blocked AI use that could aid biological weapons

Overview

Updated Sep 11

Anthropic says it blocked attempts by scientists and state-linked actors to use its Claude AI for research that could support biological weapons development. The company's third misuse report covers December 2025 through August 2026.

The structural change: Anthropic's older models were too weak to meaningfully assist dangerous biological research, so lighter safeguards sufficed. Its newer models have crossed that threshold, and the company says it can no longer make that assurance.

In one case, a scientist used Claude for gain-of-function research on the chikungunya virus, aiming to make the pathogen more transmissible and better at evading immunity. Anthropic blocked the request after learning the work would happen at a military research institute.

Why it matters

AI that accelerates vaccine research can also help build more dangerous pathogens. Anthropic says its latest models have crossed that threshold.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

Biological misuse cases blocked

Cases where Anthropic says Claude was used for research that could support biological weapons development.

Conventional weapons development cases

Attempts to use Claude to develop firearms, missiles, drones, bombs, and targeting systems.

Misuse cases involving Mythos-class models

No biological or conventional weapon misuse involved Anthropic's newest model class, except one distillation case.

Misuse reports published

This is the third in Anthropic's series of reports outlining misuse prevention.

Voices

Curated perspectives — historical figures and your fellow readers.

People Involved

Dario Amodei

Chief executive and co-founder, Anthropic

Overseeing Anthropic's safety posture amid military-use and misuse controversies

Jacob Coxon

Former Anthropic researcher

Resigned September 8, 2026

Organizations Involved

Anthropic

Artificial intelligence company

Self-regulating misuse of its Claude models; valued near $1 trillion

AI company behind the Claude model family, positioning itself as a safety-first alternative to rivals.

Timeline

December 2025 September 2026

5 events Latest: September 11th, 2026 · 2 months ago

Tap a bar to jump to that date

September 2026
Report gains global news coverage
Latest Coverage

BBC, NYT, CNBC, and other outlets highlight the bioweapons misuse angle and its implications for AI governance.
September 11th, 2026
Anthropic releases third misuse report
Report

Report details five biological misuse cases, six conventional weapons cases, and surveillance operations linked to state actors.
September 10th, 2026
Researcher Jacob Coxon resigns from Anthropic
Statement

Coxon warns the AI industry is "gambling with our lives," drawing attention days before the misuse report.
September 8th, 2026
May 2026
Scientist seeks Claude help for chikungunya gain-of-function research
Incident

An unnamed scientist uses Claude to plan experiments making the virus more transmissible, at a military research institute.
May 2026
December 2025
Anthropic begins tracking misuse cases
Monitoring

Company starts documenting attempts to use Claude for malicious activity, covering the next nine months.
December 1st, 2025

Scenarios

Congress imposes binding biosecurity rules on AI labs

Possible Resolves by Q2 2027

Discussed by: Regulators and AI safety researchers cited in the report

The report's findings feed into existing federal AI safety frameworks. A coalition of lawmakers introduces legislation requiring frontier AI labs to screen biological misuse and report incidents to a federal body. Labs that fail to comply face funding restrictions or export controls.

Misuse shifts to open-weight models

Likely Resolves by Q1 2027

Discussed by: Security researchers tracking AI misuse

As Anthropic and rivals tighten safeguards on closed models, bad actors route around them using open-weight models that lack comparable screening. The report already documents users trying third-party platforms to bypass restrictions. This scenario sees documented bioweapons-adjacent misuse of open models within a year.

Tiered verification becomes the industry norm

Possible Resolves by Q2 2027

Discussed by: AI industry observers

Anthropic's approach, restricting dual-use biological queries on capable models, gets adopted by other frontier labs. Major labs require institutional verification for high-risk biological research queries, similar to how some tools already gate access to dangerous materials.

Historical Context

3 moments from history that rhyme with this story — and how they unfolded.

September 2011 – January 2012

H5N1 gain-of-function controversy (2011-2012)

Scientists created H5N1 flu strains transmissible between ferrets, then faced a federal request to redact their methods from publication. The episode exposed the dual-use problem: research that reveals how pathogens spread can also guide bioweapon construction.

Then

A voluntary moratorium on gain-of-function flu research followed.

Now

The National Science Advisory Board for Biosecurity formalized oversight of dual-use research of concern.

Why this matters now

Same dilemma Anthropic faces: research that improves vaccines can also make pathogens more dangerous.

February 1975

Asilomar Conference on Recombinant DNA (1975)

Scientists convened to set voluntary safety guidelines after early recombinant DNA experiments raised concerns about creating new pathogens. The conference established containment levels for genetic research.

Then

Researchers adopted staged containment protocols.

Now

The conference set a precedent for scientist-led risk governance before government regulation arrived.

Why this matters now

Anthropic's self-published misuse reports are a similar voluntary move, arguably to preempt harder regulation.

July 2015

Autonomous weapons open letter (2015)

Stephen Hawking, Elon Musk, and thousands of researchers signed an open letter calling for a ban on autonomous weapons. It was one of the first major public pushes to govern emerging AI capabilities before they were fielded.

Then

Little direct policy change; the letter raised public awareness.

Now

It preceded a broader AI safety movement and helped shape labs like Anthropic itself.

Why this matters now

Shows both the limits and the reach of voluntary AI governance efforts.