AI Biosafety Risks: What the Anthropic September 2026 Threat Report Reveals, v2
By Donald Harvey Marks, MD, PhD
Physician, Microbiologist & Immunologist, 3rd Generation Veteran
1 October 2026
Anthropic’s September 2026 threat intelligence report, Detecting and Countering Misuse of AI, documents malicious campaigns disrupted across its Claude model suite between December 2025 and August 2026. The findings highlight an urgent public dilemma: the computational algorithms that accelerate vaccine discovery and rare disease therapies can also facilitate high-risk gain-of-function research and biological weaponization.
As an internal medicine physician and medical researcher with a doctorate in microbiology and immunology, military biodefense training, and decades developing recombinant vaccines, I view these disclosures not as hypothetical fears, but as urgent warnings regarding frontier biotechnology.
Key Disruptions revealed in the September 2026 Anthropic Intelligence Report
Anthropic's audit details real-world interventions across seven core threat categories: cyber operations, foreign influence, surveillance, financial fraud, biological misuse, conventional munitions design, and illicit model distillation.
The operations primarily compromised commercial tiers—including Claude Haiku, Sonnet, and Opus—while higher-tier frontier systems remained secured. Key state and non-state interventions included:
Transnational Surveillance Campaigns: A PRC-aligned intelligence cell utilized Claude to compile detailed, Chinese-language intelligence dossiers targeting Tibetan Buddhist civil networks and the Central Tibetan Administration.
Middle East Military Targeting: An operative linked to Iranian interests attempted to extract operational targeting recommendations against U.S. naval formations deployed in the Middle East.
Industrial-Scale Model Distillation: Adversarial entities and rival overseas developers (including DeepSeek, Moonshot AI, and MiniMax) systematically executed extraction protocols against Claude’s internal weights.
State-Backed Pathogen Queries: Multiple actors deployed coordinated queries seeking technical assistance with viral pathogens, toxins, and infrastructure exploitation.
The Dual-Use Dilemma: Medicine vs. Bioweapons
The central challenge in frontier biosafety is the inescapable dual-use dilemma: the exact genetic mapping required to formulate protective vaccines can be mirrored to engineer enhanced pathogenicity.
Anthropic flagged five independent incidents where researchers prompted Claude for assistance with workflows applicable to biological munitions. Simultaneously, the platform verified that the queries carried identical utility for standard antiviral discovery.
Because intent is obscured during initial prompt intake, automated classifiers struggle to distinguish between a laboratory preparing a prophylactic response and a hostile team engineering pandemic vectors.
Case Analysis: Chikungunya Gain-of-Function Research
The most alarming finding in the report details an operator attempting to generate grant documentation and experimental protocols for gain-of-function engineering on the Chikungunya virus.
Protocol Objectives: The submitted request directed the AI to model mutations that increase Chikungunya's airborne transmissibility and enhance immune-evasion properties against standard antibodies.
Gain-of-Function Defined: This represents the clinical definition of high-risk gain-of-function: genetically modifying a pathogen to introduce or amplify a dangerous biological capability.
Institutional Ties: While the interaction involved academic researchers, intelligence traces connected the request directly to a foreign military research center.
When an AI system calculates the evolutionary biology needed to make a crippling alphavirus more durable and transmissible, the line separating laboratory defense from biowarfare ceases to exist.
Pathogen Weaponization: From Bird Flu to Aerosolized Agents
Beyond Chikungunya, the Anthropic audit blocked queries involving highly pathogenic avian influenza (HPAI / bird flu) and other classified CBRN toxins.
Anthropic declined to identify the specific academic institutions or foreign nations submitting these prompts, citing inconclusive evidence regarding end-user intent. However, from an operational biodefense perspective, attempts to optimize the thermal stability, environmental persistence, and lethality of respiratory pathogens point toward weapons exploration.
These findings align with concerns raised in modern national security literature, including Annie Jacobsen’s analysis of modern biological warfare vulnerabilities.
Lowering the Expertise Barrier: Why Safeguards (ASL-3) Are Vital
The core revelation of the September 2026 report is that artificial intelligence is compressing the expertise barrier for biological synthesis.
Historically, weaponizing a pathogen required specialized institutional expertise, complex laboratory supply chains, and years of trial-and-error biology. Large language models lower this technical barrier by diagnosing synthesis errors, optimizing growth protocols, and troubleshooting aerosol delivery mechanics in minutes.
In response, Anthropic activated AI Safety Level 3 (ASL-3) protocols specifically engineered to restrict Chemical, Biological, Radiological, and Nuclear (CBRN) intelligence. Yet private company guardrails solve only part of the problem. Open-weights models developed overseas operate without comparable oversight, leaving high-risk life sciences tools open to rogue actors.
The Policy Response: Federal Bans and Historical Parallels
Federal policies increasingly recognize the catastrophic potential of uncontrolled life sciences research.
In 2026, the Department of Health and Human Services (HHS) enacted the U.S. Government Policy for Stopping High-Risk Life Sciences Research, establishing firm regulatory boundaries:
Funding Prohibitions: A ban on domestic and global taxpayer funding for high-risk gain-of-function research on hazardous pathogens.
Stringent Definitions: Explicit targeting of research designed to enhance transmissibility, lethality, or immune evasion.
Mandated Oversight: Strict, top-down risk-assessment frameworks for institutional biological studies.
This directive reflects the historic precedent established in 1969 by President Richard Nixon, who unilaterally halted the U.S. offensive biological weapons program via National Security Decision Memorandum 35. Nixon dismantled offensive biological stockpiles and limited federal research strictly to defensive vaccines, diagnostics, and personal protective countermeasures. Today's frontier AI models require that same clarity of separation between defensive innovation and dangerous enhancement.
Conclusions: Guarding the Gateways of Biology
The Anthropic September 2026 report demonstrates that artificial intelligence has become an active participant in biological research. While these tools offer promising advances in vaccine manufacturing and clinical care, they simultaneously offer an algorithmic blueprint for engineering disease.
Voluntary corporate filtering is an insufficient safeguard. Protecting civil society against deliberate or accidental biological catastrophe requires international enforcement, verified safety layers, and absolute transparency from private AI laboratories before higher-tier models are released into the wild.
Author Perspective & Qualifications
Donald H. Marks, MD, PhD: Earned both an MD and a PhD in Microbiology and Immunology from UCLA.
Research & Development: Principal investigator on 5 NIH grant proposals and 1 DARPA project; co-inventor on multiple patented recombinant vaccine technologies.
Military & Biodefense: Former active-duty military medical officer and clinical researcher at the Letterman Army Institute of Research (LAIR), with biological warfare training completed at Fort Detrick (USAMRIID), Maryland.
Pharmaceutical Industry: Former clinical vaccine development leader with Aventis Pasteur and Hoffman-La Roche.
References & Foundational Studies
1. Anthropic report 2026 “Detecting and countering the misuse of AI.
2. Trump ends dangerous gain of function research. Karie Pavlich, Townhall 19 June 2025
4. The September 2026 anthropic report on AI dangers - what it means. DH Marks 18 Sept 2026
5. Anthropic quietly sets up biology lab as it ramps AI drug program - Reuters 18 Sept 2026
6. Claude discovers a novel enzyme system with CRISPR-like repeats. Sep 23, 2026.
7. Keller, D, Koster, FT, Marks, DH et al. Safety and Immunogenicity of a Recombinant Outer Surface Protein A Lyme Vaccine. JAMA, June 8, 1994, p 1764-1768.
8. Sigal LH --- Marks D, Malawista SE. A vaccine consisting of recombinant Borrelia burgdorferi outer surface protein A to prevent Lyme disease. Recombinant Outer-Surface Protein A. Lyme Disease Vaccine Study Consortium. N Engl J Med (1998 Jul 23) 339(4):216-22.
9. Kanesa-thasan N --- Marks DH et al. Safety and immunogenicity of NYVAC-JEV and ALVAC-JEV attenuated recombinant Japanese encephalitis virus--poxvirus vaccines in vaccinia-nonimmune and vaccinia-immune humans. Vaccine (2000 Oct 15) 19(4-5):483-91.
10. VanKampen KR --- Marks D, Elmets CA, and Tang DC. 2004. Safety and immunogenicity of adenovirus-vectored nasal and epicutaneous influenza vaccines in humans. Vaccine. 2005 Jan 11;23(8):1029-36.
11. Jonas Salk, Polio Vaccine and Vaccinating Against Hate. The healthcare discovery that saved the lives of millions of children. Remembrance of my afternoon in Paris with vaccine titan Jonas Salk. by Donald H Marks
12. What is known about the memo from Dr Fauci on Ebola research at USAMRID
