Anthropic Reports Blocking AI Misuse for Biological Weapons, Cyberespionage, and Missile Development

Anthropic published a threat-intelligence report on 10 September 2026 documenting dozens of cases in which its Claude models were misused for cyberespionage, surveillance, biological research, and conventional weapons development between December 2025 and August 2026, with the responsible accounts banned. The report details five biological cases involving chikungunya mutations, avian flu transmissibility, smallpox resistance, and toxin catalogues, a Russian-linked campaign consistent with Midnight Blizzard targeting more than two dozen Ukrainian government organizations, and a Yemen-based cell that used Claude to write guidance, navigation, and control software for a guided rocket and a multi-stage ballistic missile with a target range above 2,000 km. Anthropic head of threat intelligence Jacob Klein said it remains unknown whether the biological research was intended to be weaponised, describing the situation as "incredibly nuanced.

Anthropic published a threat-intelligence report on 10 September 2026 documenting dozens of cases in which its Claude models were misused for cyberespionage, surveillance, biological research, and conventional weapons development between December 2025 and August 2026, with the responsible accounts banned <sup 3 </sup <sup 6 </sup <sup \ 7\ </sup src-007 . The report describes five biological cases in which users sought help with research on chikungunya mutations, avian flu transmissibility, smallpox resistance, and toxin catalogues, including a scientist at a military research institute seeking grant-writing assistance to engineer more harmful chikungunya variants <sup 3 </sup <sup 12 </sup <sup \ 14\ </sup src-014 . Anthropic's head of threat intelligence, Jacob Klein, said it remains unknown whether the biological research was intended to be weaponised, describing the situation as incredibly nuanced <sup 3 </sup <sup \ 7\ </sup src-007 . The report also documents a Russian-linked cyber-espionage campaign attributed to a group consistent with Midnight Blizzard that targeted more than two dozen Ukrainian government organizations, hijacked WhatsApp accounts of former senior Ukrainian officials, and auto-rewrote malware to evade detection <sup 6 </sup <sup 19 </sup <sup \ 36\ </sup src-036 . A Yemen-based cell, identified by the Financial Times as likely Houthi, used Claude to write guidance, navigation, and control software for a guided rocket, a multi-stage ballistic missile with a target range above 2,000 km, and a hypersonic glide-vehicle variant, though Anthropic says no operational device was deployed <sup 20 </sup <sup 24 </sup <sup \ 39\ </sup src-039 ; a Yemeni outlet cautioned that attributing the cell to the Houthis requires independent evidence <sup \ 40\ </sup src-040 . State-sponsored surveillance operations originating in China, Iran, and Mali targeted Hong Kong pro-democracy activists, Tibetan and Falun Gong communities, and Iranian diaspora opponents, with a contractor for Mali's national security authorities designing a platform capable of monitoring 25 million mobile subscribers <sup 14 </sup <sup 29 </sup <sup \ 44\ </sup src-044 . Chinese developers Moonshot AI and DeepSeek were accused of routing customer queries through Claude via networks of fraudulent accounts to distill its capabilities, with Alibaba-linked operators generating over 151 million interactions over three months <sup 6 </sup <sup 29 </sup <sup \ 42\ </sup src-042 . The report's publication coincided with the resignation of former Anthropic researcher Jacob Coxon, who accused Anthropic and OpenAI of "gambling with our lives" in a race toward self-improving superintelligence that could kill everyone by the end of the decade <sup 4 </sup <sup 10 </sup <sup \ 27\ </sup src-027 . Anthropic's Alignment Science lead, Evan Hubinger, publicly endorsed Coxon's concerns, stating that he personally estimates the probability of AI-caused human extinction at more than 10 percent within ten years and that no complete plan exists to align a future superintelligence <sup 9 </sup <sup 17 </sup <sup \ 32\ </sup src-032 . Berkeley computer science professor Stuart Russell warned that an out-of-control superintelligent AI could synthesise novel pathogens or trick humans into nuclear war, and urged a licensing regime akin to aviation or medicine <sup \ 10\ </sup src-010 . Senator Bernie Sanders called for a pause on advanced AI and a ban on superintelligence, saying, "When scientists tell you there is a chance, a chance that it could have a cataclysmic impact on humanity, you've got be a moron not to say, slow it down" <sup \ 7\ </sup src-007 . UN High Commissioner for Human Rights Volker Türk told the Human Rights Council that steps must be taken before AI becomes a threat to humanity's existence, urging states to set limits and ban weapons that can take lives without direct human involvement <sup \ 23\ </sup src-023 . Other scientists and executives rejected the extinction framing. UNSW Canberra computing professor Hussein Abbass called percentage-based doom predictions "absolutely unscientific" and said the situation remains controllable and manageable <sup \ 10\ </sup src-010 . Armin Grunwald, head of the Office of Technology Assessment at the German Bundestag, said AI could cause "truly very serious things" but he sees "keine Auslöschung der Menschheit" no extinction of humanity , and urged focus on other dangers such as people believing they no longer need to learn or develop basic human capabilities <sup \ 25\ </sup src-025 . Replit CEO Amjad Masad disputed the intelligence-explosion premise, predicting AI agents will not be much better than humans and that many bottlenecks on intelligence will take years to solve <sup \ 9\ </sup src-009 . AI skeptic Gary Marcus argued that a decade of apocalyptic discourse has mainly served to make AI companies richer and more powerful <sup \ 32\ </sup src-032 . President Donald Trump dismissed concerns about AI causing human extinction, saying, "No, I don't have any," and argued the US must not restrict development lest it "come in second to China" <sup 3 </sup <sup 7 </sup <sup \ 10\ </sup src-010 . The governance debate centered on whether voluntary corporate safeguards can substitute for binding government regulation. OpenAI reversed its earlier opposition to tighter oversight, with chief global affairs officer Chris Lehane calling for mandatory federal safety requirements on the most powerful AI systems, including standardized testing, independent evaluations, and mandatory incident reporting <sup \ 4\ </sup src-004 . Johns Hopkins Center for Health Security director Tom Inglesby credited Anthropic for catching and publishing the biological misuse attempts but warned that "there is no US government policy requiring frontier models or biological AI models to be assessed for major vulnerabilities" or for national-security biological risks <sup \ 34\ </sup src-034 . Der Spiegel argued in an editorial that the AI industry warns about its products' dangers while refusing to forgo the business, so the only remedy is to force companies to exercise more caution <sup \ 30\ </sup src-030 . Three experts told Politico Europe that the first step is for laboratories, governments, and international bodies to agree on common testing standards <sup \ 22\ </sup src-022 . Anthropic itself has refused to let a British AI security agency test its latest model, according to Bloomberg Opinion <sup \ 9\ </sup src-009 . Biodefense scholar Gregory Koblentz of George Mason University pushed back on Anthropic's characterization of the blocked biological research as dangerous dual-use work, arguing it is better described as "multi-use" science with legitimate peaceful, defensive, and commercial applications, and that studying endemic pathogens or having military affiliations does not by itself indicate a state weapons program <sup \ 34\ </sup src-034 . Council on Strategic Risks researcher Andrew Weber took the opposite position, calling the findings "chillingly realistic" evidence that state-backed biological weapons developers are exploiting rapidly advancing AI, and urging that model access be limited to trusted researchers <sup 3 </sup <sup 34 </sup <sup \ 46\ </sup src-046 . China's foreign ministry spokesperson Mao Ning rejected the report as a distortion of facts and a smear against China, saying China "has always advocated the development of AI for good" <sup \ 29\ </sup src-029 . China's Ministry of Commerce called the distillation accusations "于事无凭，于法无据" baseless in fact and law and accused the US of politicizing a normal industry technique <sup \ 42\ </sup src-042 . Russian foreign ministry spokesperson Maria Zakharova said no concrete evidence has ever been presented for claims of Russian malicious cyber activity, calling the accusations hypocritical and casting Russia as the real victim of Western cyber operations <sup \ 37\ </sup src-037 . Reporting from Anadolu Agency and Al Jazeera noted that Anthropic's misuse report omits its own military relationship with the Pentagon, including a collapsed $200 million contract, the White House's move to phase Anthropic out of federal use after it refused to lift safeguards on autonomous weaponry, and claims that Claude has nonetheless been deployed in military missions in Iran and Venezuela <sup 6 </sup <sup \ 26\ </sup src-026 . Anthropic CEO Dario Amodei has framed pursuing industry preeminence as a way to "pull the ecosystem along with us," without spelling out how racing on development makes AI safer <sup \ 9\ </sup src-009 . Anthropic also removed from its safety charter a pledge to halt development if it failed to control model risks, arguing a unilateral pause would let less cautious rivals dominate <sup \ 10\ </sup src-010 . Commercial momentum continued alongside the safety debate. Reuters reported that Nvidia is in preliminary talks to anchor-invest in Anthropic's anticipated multibillion-dollar IPO <sup \ 35\ </sup src-035 . In Thailand, AIS Academy staged a seminar drawing over 6,000 registrations under the theme "Mastering AI Beyond Borders," with enterprise chief Kantima Lerlertyuttitham framing the event as moving Thailand from generative AI toward AGI capability <sup \ 18\ </sup src-018 . OpenAI CEO Sam Altman told employees the company could moderate its pace of developing cutting-edge models and hoped other AI laboratories would coordinate, though he noted some competitors might refuse to join <sup 1 </sup <sup \ 28\ </sup src-028 . OpenAI had already halted training of its latest models for two weeks in August before resuming under tighter controls <sup \ 10\ </sup src-010 , and delayed parts of its GPT-6 Astra model after tests revealed advanced cybersecurity capabilities <sup \ 4\ </sup src-004 . US Senator Bernie Sanders and Representative Greg Casar introduced bills to halt superintelligence, while UK MP Alex Sobel introduced a similar AI safety bill treating recursive self-improvement as a precursor requiring regulation <sup \ 21\ </sup src-021 . A UK open letter has sought a multilateral treaty on AI safety <sup \ 45\ </sup src-045 . Anthropic says it shared threat information with public and private partners and engaged an independent research firm to review the incidents, and that it disrupted every operation covered in the report <sup 20 </sup <sup \ 26\ </sup src-026 .