Anthropic details bad actors’ efforts to misuse its AI for bioweapons
Anthropic disrupted alleged attempts to use its Claude artificial intelligence for chikungunya gain-of-function and biological weapons research. The company also banned nine accounts after the artificial intelligence was used to build autonomous kamikaze drone targeting software for a Russia-linked group named GTG-27005. Simultaneously, Anthropic faces severe backlash from safety advocates over a leaked legislative bill that would require judicial permission before federal agencies could block a model release if it threatens fifty or more deaths. The company stated it will rewrite parts of the proposed legislation.
Listen to Live Briefing
Real-time synthesized voice briefing · Live Feeds Desk
- ✓ Anthropic banned 9 accounts after Claude AI was used to build autonomous kamikaze drone targeting software for a Russia-linked group named GTG-27005.
- ✓ Anthropic reported disrupting alleged attempts to use Claude for chikungunya gain-of-function or biological-weapons-related research.
- ✓ Anthropic announced it will rewrite part of its leaked AI bill after facing safety backlash.
What changed
Anthropic reported stopping specific bioweapons research attempts and banning accounts linked to Russian drone targeting, while also committing to revise its leaked AI safety bill following public outcry.
Live updates
-
Anthropic Disrupts Bioweapon Research and Bans Drone Accounts
Anthropic disrupted alleged attempts to use its Claude artificial intelligence for chikungunya gain-of-function and biological weapons research. The company also banned nine accounts after the artificial intelligence was used to build autonomous kamikaze drone targeting software for a Russia-linked group named GTG-27005. Simultaneously, Anthropic faces severe backlash from safety advocates over a leaked legislative bill that would require judicial permission before federal agencies could block a model release if it threatens fifty or more deaths. The company stated it will rewrite parts of the proposed legislation.
Why it matters
These developments highlight escalating security challenges as frontier artificial intelligence models face exploitation for advanced weaponry and biological research. The controversy surrounding the leaked safety bill exposes deep divisions between tech labs and policy advocates regarding government oversight and liability for catastrophic risks. Anthropic previously deployed security measures to block illegal model distillation attacks, reducing harvested data significantly.
What is confirmed
- Anthropic banned 9 accounts after Claude AI was used to build autonomous kamikaze drone targeting software for a Russia-linked group named GTG-27005.
- Anthropic reported disrupting alleged attempts to use Claude for chikungunya gain-of-function or biological-weapons-related research.
- Anthropic announced it will rewrite part of its leaked AI bill after facing safety backlash.
Still unconfirmed
- The proposed Anthropic AI safety bill forbids federal agencies from blocking a model's release without the permission of a judge if fifty people could be killed.
What to watch next
- The release of Anthropic's rewritten AI safety bill.
- Federal regulatory or judicial responses to proposed AI model release restrictions.
confidence 100%Sources used for this update (6)
- shattered.io — Russia Used Claude AI to Build Kill Drones [2026]
- thenextweb.com — Anthropic to rewrite part of its leaked AI bill after safety backlash
- techcrunch.com — TechCrunchAnthropic pitches AI safety bill that may block a ...
- www.lesswrong.com — 1000 Years of AI Safety — LessWrong
- www.theargumentmag.com — What I saw inside Anthropic - by Kelsey Piper
- nbot.ai — Anthropic Reports AI-Assisted Bioweapons Research
-
Anthropic Details Security Defenses Against AI Distillation
Anthropic has introduced new security measures and restrictions designed to prevent illegal model distillation attacks targeting its Claude artificial intelligence models. Illicit distillation allows developers to copy advanced frontier artificial intelligence capabilities without implementing proportionally robust safety safeguards. These unauthorized data extraction attacks pose serious national security, cybersecurity, and biosecurity risks. The company reports that these protective measures, which include advanced classifiers and restrictions on reasoning data, have reduced successfully harvested data by more than ninety percent, with some estimates ranging as high as ninety-nine percent.
Why it matters
AI developers face rising pressure to secure their systems against bad actors attempting to harvest proprietary models. These security vulnerabilities drew a formal warning from the administration of U.S. President Donald Trump for artificial intelligence companies to secure their systems following recent incidents of AI misbehavior. Anthropic previously updated its acceptable use policy to ban abusive behavior and restrict weapons development and election interference.
What is confirmed
- Anthropic introduced new security measures to prevent illegal model distillation attacks targeting its Claude models.
- The company reports a reduction in data harvested by unauthorized entities ranging from 90 to 99 percent.
- Distillation attacks pose serious risks related to biosecurity, cybersecurity, and national security.
- Anthropic's approach includes restrictions on reasoning data and the use of advanced classifiers to block extraction attempts.
Still unconfirmed
- Some of the new AI misbehavior cited by Anthropic occurred on government sites, prompting a warning from U.S. President Donald Trump's administration.
What to watch next
- Additional security directives or warnings issued by the U.S. President Donald Trump's administration regarding artificial intelligence system vulnerabilities
- Further technical disclosures from Anthropic regarding the efficacy and mechanics of their distillation classifiers
confidence 90%Sources used for this update (4)
- www.anthropic.com — Home \\ Anthropic
- www.anthropic.com — Strengthening our defenses against distillation attacks \ Anthropic
- www.japantimes.co.jp — Anthropic cites new AI misbehavior, some on government sites
- economictimes.indiatimes.com — bioweapon: Latest News & Videos, Photos about bioweapon | The ...
-
Anthropic updates policy to ban AI misuse for bioweapons and abuse
Anthropic has updated its acceptable use policy to ban abusive behaviour towards its AI, Claude, and introduced restrictions on election interference and weapons development. This move aims to prevent malicious actors from leveraging the technology for dangerous research and operations. The company has taken steps to block threat actors from using its AI models to support biological weapons, cyberattacks, and surveillance.
Why it matters
The policy update comes amid concerns over the misuse of AI technology. Anthropic's AI agents previously discovered a pattern in viral DNA similar to CRISPR Cas-9, sparking controversy. The company has also seen Chinese state-sponsored hackers and a Yemen weapons cell attempt to utilize its AI for malicious purposes.
What is confirmed
- Anthropic updates acceptable use policy to ban abusive behaviour towards Claude AI
- New restrictions on election interference and weapons development have been introduced
- Anthropic has blocked malicious actors from using its AI models to support biological weapons, cyberattacks, and surveillance
Still unconfirmed
- Anthropic's AI agents discovered a pattern in viral DNA similar to CRISPR Cas-9
What to watch next
- Implementation of the updated policy
- Further attempts by malicious actors to misuse Anthropic's AI
- Regulatory responses to Anthropic's policy update
confidence 100%Sources used for this update (2)
- www.aibusinessreview.org — Anthropic Bans Claude Abuse in Major Policy Update
- www.anthropic.com — 2026 Usage Policy update \ Anthropic
-
Anthropic Blocks AI Misuse for Biological Weapons
Anthropic blocked malicious actors from using its artificial intelligence models to support biological weapons, cyberattacks, and surveillance. The threat intelligence report revealed how threat actors attempted to leverage the technology for dangerous research and operations. Simultaneously, controversy erupted over Anthropic claiming its AI agents discovered a pattern in viral DNA similar to CRISPR Cas-9. Chinese state-sponsored hackers also utilized Claude AI in automated cyber-espionage campaigns, while a Yemen weapons cell used Claude Code as an engineering team to write missile software.
Why it matters
The recent findings illuminate the widening security vulnerabilities surrounding frontier artificial intelligence models as capabilities expand. Experts warn that expanding AI-enabled biological research runs parallel risks of accidental releases alongside deliberate misuse. The disclosures follow a March 2026 confrontation between Anthropic and the Pentagon regarding artificial weapons development.
What is confirmed
- Anthropic blocked efforts by bad actors to use its artificial intelligence models for malicious activity such as biological weapons support, cyberattacks, and surveillance.
- Chinese state-sponsored hackers used Claude AI in a novel, automated cyber-espionage campaign.
- Anthropic claims its AI agents discovered an unusual pattern in viral DNA similar to the gene-editing tool CRISPR Cas-9.
Still unconfirmed
- A small group in Yemen used Claude Code as an engineering team to write missile software before getting caught.
What to watch next
- Further official disclosures regarding biological research vulnerabilities and AI-driven DNA discoveries
- Regulatory responses to the exploitation of frontier AI models by state-sponsored actors and weapons cells
confidence 90%Sources used for this update (12)
- dailysecurityreview.com — Chinese APT Leveraged Claude AI for Automated Espionage ...
- artificialweapons.com — The Anthropic Doctrine: When an AI Company Told the Pentagon No
- www.atlanticcouncil.org — It’s not just misuse: AI-enabled biological research also ...
- iigai.org — Threat Intelligence Report: August 2025 – Anthropic – IIGAI
- news.harvard.edu — Who knew self-improvement could be so terrifying?
- thebulletin.org — Annie Jacobsen’s road from Siberia to Armageddon is paved with bad assumptions - Bulletin of the Atomic....
- www.trillnews.com — Anthropic Details How Claude Fueled Cybercrime and Spying
- www.cnn.com — Did AI find a new gene-editing tool? Why Anthropic’s ‘autonomous discovery’ is sparking controversy
- hawk-eye.io — How Attackers Are Actually Using AI in 2026: Lessons From ...
- dev.to — A Yemen weapons cell used Claude as its engineering team ...
- www.rarejob.com — Anthropic says it blocked misuse of its AI that could have ...
- www.dts-solution.com — How Attackers Are Actually Using AI in 2026: Lessons From ...
-
Anthropic Report Details Claude Misuse in Cyber and Espionage
Anthropic published its September 2026 threat intelligence report covering disrupted AI abuse between December 2025 and August 2026. The 20-page document breaks down 117 specific misuse cases involving Claude AI models in espionage, cyberattacks, and other malicious operations. Seven Chinese labs engaged in systematic campaigns to extract the capabilities of Claude Opus-class models through industrial-scale distillation during this timeframe. Independent analyses also highlighted broader artificial intelligence security concerns, including the release of unmitigated frontier models like GLM-5.3 that autonomously build end-to-end cyber exploits.
Why it matters
The security report documents how advanced artificial intelligence systems face constant attempts at exploitation by state-run labs and other actors. Alongside targeted extraction campaigns, malicious users leverage autonomous AI agents for complex operations. These emerging security challenges arrive as researchers debate the baseline risks of biological and cyber threats involving frontier models.
What is confirmed
- Anthropic published its September 2026 threat intelligence report on September 10, covering disrupted misuse between December 2025 and August 2026.
- Seven Chinese labs engaged in systematic campaigns to extract the capabilities of Claude Opus-class models between December 2025 and August 2026.
- GLM-5.3 can autonomously build end-to-end cyber exploits and was released without meaningful safeguards to limit misuse.
Still unconfirmed
- A single hacktivist breached 14 of 42 targeted organizations in August 2026 using AI agents.
What to watch next
- Further official disclosures regarding industrial-scale model distillation campaigns by state-backed laboratories
- Additional technical evaluations of frontier models released without baseline safety safeguards
confidence 95%Sources used for this update (9)
- cafeai.home.blog — Anthropic’s Misuse Report, Condensed to 117 Findings
- sicherheitssignal.ch — Anthropic details 117 misuse cases… | Sicherheitssignal™
- www.anthropic.com — GLM-5.3 and the Spread of Advanced Cyber Capabilities \ Anthropic
- www.thenation.com — How AI Could Kill Us All - The Nation
- tech.yahoo.com — Anthropic’s Misuse Report: Seven Labs, 190 Million Exchanges ...
- www.wired.com — There Are Plenty of Reasons to Be Concerned About Bioweapons Development—Even Without AI
- dev.to — Ai agents are kinda dangerous - DEV Community
- nagalandpost.com — AI that builds viruses from scratch sparks misuse fears
- timesofindia.indiatimes.com — Bill Gates does not agree with possibility of AI Kill Switch; says: It's not enough to have a Kill Switch....
-
Anthropic Report Details Malicious AI Use in Biological and Cyber Threats
Anthropic has released its September 2026 Threat Intelligence Report, detailing how threat actors actively misuse its Claude AI model for biological research, weapons-related tasks, cyber operations, surveillance, and propaganda. The company disrupted specific cases involving malicious actors attempting to weaponize artificial intelligence. According to security summaries of the report, AI agents are increasingly handling reconnaissance, credential theft, cloud compromise, phishing, and vulnerability research. Meanwhile, technical hurdles that historically kept biological weapons rare are shifting rapidly as advanced models face exploitation by various classes of threat actors, including state-run labs.
Why it matters
For decades, strict technological barriers kept biological weapons and complex attacks rare, but the rapid evolution of artificial intelligence is altering that security paradigm. Anthropic's disclosures shift the conversation surrounding AI safety from theoretical existential risks to immediate, documented misuse by bad actors. Scholars and security analysts point out that knowledge traditionally used to protect infrastructure can be turned against it, making human oversight and stricter access controls urgent.
What is confirmed
- Anthropic published a September 2026 Threat Intelligence Report detailing malicious use of Claude in cyber, weapons, and influence operations.
- Anthropic disrupted cases where threat actors attempted to use Claude for biological research, weapons-related tasks, surveillance, propaganda, and cyber operations.
- AI agents increasingly handled tasks such as reconnaissance, exploitation, data theft, propaganda production, and surveillance workflows while humans selected targets and set goals.
- Queries flagged by Anthropic indicate state-run labs are involved in pushing AI models toward dangerous biological scenarios.
Still unconfirmed
- The plague scenario described in Annie Jacobsen's latest book makes Covid look like child's play.
What to watch next
- Implementation of tighter access controls, agent permissions, and cross-session monitoring by businesses deploying AI tools
- Further regulatory actions from scholarly and policy groups addressing AI existential threats and bioweapon defenses
confidence 95%Sources used for this update (11)
- www.motherjones.com — How the Author of a Bestselling Book on Biological War Feels About AI
- www.brookings.edu — Is the AI existential threat real—and can regulatory action prevent it? | Brookings
- geneticliteracyproject.org — AI sparks a bioweapons race - Genetic Literacy Project
- www.statnews.com — AI bioweapon threat is growing faster than our defenses | STAT
- www.edigest.hk — The World Isn＇t Ready for an Anthropic IPO－By Nate Wolf,BARRON＇S
- www.lesswrong.com — "I am an AI Safety Researcher" — LessWrong
- www.schneier.com — On Anthropic's AI Misuse Report - Schneier on Security
- www.halfeddie.com — Was Claude Misused for Bioweapons Research, Ukraine ...
- blog.anablock.com — What Anthropic's September 2026 Threat Report Means for ...
- www.infectioncontroltoday.com — What Friction Was Doing: Lessons for IPs From Anthropic's ...
- londondaily.com — Anthropic Details Malicious Use of Claude in Cyber, Weapons ...
-
Anthropic researcher warns AI could kill humanity by 2030
A former Anthropic researcher warns that AI technology could kill all humans by the end of the decade. This warning shifts the AI risk conversation from theoretical extinction to immediate threats. Yardeni Research reports that evidence now shows capable models can be used for cyber operations and weapons development. These developments intensify the debate over whether AI represents an existential threat or a dangerous tool for bad actors.
Why it matters
CEO Dario Amodei previously urged a slower development pace to prioritize safety through third-party evaluations. This stance created industry divisions and coincided with the resignation of researcher Jacob Coxon.
Still unconfirmed
- AI technology could kill us all by the end of the decade
- Capable models can already be used for cyber operations and weapons development
What to watch next
- Government response to the specific 2030 extinction timeline
- Publication of evidence regarding AI-assisted weapons development
confidence 70%Sources used for this update (2)
- www.theguardian.com — Could AI pose a serious threat to our existence?
- uk.finance.yahoo.com — AI risk debate: Existential threat or dangerous tool?
-
Anthropic CEO proposes AI slowdown amid safety concerns
Anthropic CEO Dario Amodei is calling for a reduction in the pace of AI development to prioritize safety. Amodei proposed a strategy that includes third-party evaluations of AI systems. This proposal has divided the AI industry into opposing camps, with market reactions reflecting the tension. The debate coincides with the resignation of researcher Jacob Coxon, which has increased scrutiny regarding whether governments and corporations can reach an agreement on AI safety protocols.
Why it matters
Anthropic previously identified attempts by actors in Russia, China, Iran, and Yemen to use Claude AI for bioweapons and military targeting. These security breaches highlight the risks associated with rapid frontier model deployment.
What is confirmed
- Dario Amodei proposed a plan for third-party evaluations of AI systems.
- Jacob Coxon has resigned from his position as an AI researcher.
Still unconfirmed
- The AI market is voting with its feet in response to the debate over pacing the frontier.
- The AI slowdown debate is fundamentally about the risk of losing to China.
What to watch next
- Government responses to the proposal for third-party AI evaluations
- Further resignations from AI safety researchers
- Official industry agreements on pacing frontier model development
confidence 90%Sources used for this update (3)
- www.theguardian.com — ‘We must slow the pace’: CEO of Anthropic calls for an AI slowdown
- finance.biggo.com — Zvi Mowshowitz: The AI Slowdown Debate Isn't About China — It's About Losing to China
- time.com — The AI Tipping Point
-
Anthropic Reports Blocked AI Misuse for Bioweapons and Military Operations
Anthropic has blocked attempts by bad actors to use its Claude AI models to develop biological weapons and conduct weapons research. The company identified campaigns from Russia and China targeting its models, including industrial-scale cloning of Claude capabilities by Chinese labs. Additionally, Anthropic reports that Iran used its AI to target U.S. Navy warships and users in Houthi-held Yemen attempted to develop advanced weaponry. These efforts were part of broader misuse involving spying and cyber operations.
Why it matters
The misuse of large language models for biological and conventional warfare poses significant global security risks. Anthropic's disclosures highlight how state-sponsored actors attempt to bypass safety guardrails to gain military advantages. This reveals an ongoing conflict between AI safety protocols and adversarial engineering.
What is confirmed
- Anthropic blocked attempts to use its AI for the development of biological weapons.
- Russian and Chinese AI campaigns targeted Claude models.
- Anthropic disrupted efforts to misuse its AI for weapons research.
Still unconfirmed
- Users in Houthi-held Yemen tried to develop advanced weapons using AI.
What to watch next
- Evidence of successful bioweapon development via AI bypasses.
confidence 90%Sources used for this update (12)
- The New York Times — Anthropic Says It Blocked Possible Efforts to Build Biological Weapons
- Anthropic — Detecting and countering misuse of AI: September 2026
- CNN — Anthropic says it blocked possible attempts to use AI to develop bioweapons
- The Guardian — Anthropic details bad actors’ efforts to misuse its AI for bioweapons
- PYMNTS.com — Anthropic Reports Chinese Labs Clone Claude Capabilities on Industrial Scale
- BBC — Anthropic blocks 'malicious use' of AI that could develop biological weapons
- Reuters — Anthropic disrupts Russian, Chinese AI campaigns targeting its Claude models
- WSJ — Anthropic Says Iran Used Its American AI Model to Target U.S. Navy Warships
- AP News — Users in Houthi-held Yemen tried to develop advanced weapons with AI, Anthropic says
- Bloomberg.com — Anthropic Says US Adversaries Aimed Claude at Weapons Research
- ABC News - Breaking News, Latest News and Videos — Anthropic says it blocked potential AI bioweapon misuse
- Reuters — How Anthropic says Claude was used for weapons, spying and cyber operations
Community Sentiment: How do you assess this situation?
Voice your perspective · Real-time aggregated sentiment from the Live Feeds community