11-09-2026
Anthropic says it detected and disrupted attempts to misuse its Claude AI models in ways that could support the development of biological weapons. Its latest threat intelligence report describes five cases involving biological research that raised concern, while stressing that the same scientific information can have legitimate uses, such as developing vaccines or disease treatments. The company characterized biological misuse as one of the most serious risks posed by frontier AI and warned that inadequate safeguards could have catastrophic consequences.
The report covers activity detected between December 2025 and August 2026. It also details the use of Claude in cyber-espionage, surveillance, scams, fraud, propaganda, influence operations and the development of conventional weapons, including missiles, armed drones, bombs and their targeting systems. Groups and actors linked to Russia, China and Iran are mentioned, as are criminal organizations, spyware vendors and politically motivated individuals. Anthropic also accused Chinese AI companies of attempting to replicate Claude’s capabilities.
The findings are part of a broader series of industry warnings about the risks of increasingly capable AI systems. Google recently reported an attempt to use Gemini to obtain instructions for synthesizing weaponized biological agents. Anthropic’s report also follows warnings from its own researchers and OpenAI chief scientist Jakub Pachocki that AI development may be advancing faster than safety measures. One Anthropic researcher estimated that AI could have more than a 10% chance of killing all humans within the next decade.
The warnings have fueled demands for stronger regulation. US Senator Bernie Sanders has proposed pausing advanced AI development and banning artificial superintelligence, while calls have been made for an international treaty. President Donald Trump has rejected the idea of slowing AI development, arguing that the US must win the AI race. Anthropic says it has shared intelligence with authorities and industry partners and is incorporating the findings into improved safeguards.
Entities: Anthropic, Claude AI models, Biological weapons development, Frontier AI safety and safeguards, Google Gemini • Tone: urgent • Sentiment: negative • Intent: inform
11-09-2026
Anthropic says it identified and disrupted large-scale efforts by Chinese AI companies Alibaba, Moonshot AI and DeepSeek to use outputs from its Claude models to train competing systems without authorization. The company characterizes this practice as “illicit distillation,” in which responses from a more capable model are collected and used to reproduce or improve another model’s capabilities.
According to Anthropic’s threat intelligence report, Alibaba’s operation was the largest campaign detected. Between May and July, operators allegedly generated more than 151 million exchanges with Claude through over 3,500 fraudulent accounts, with activity reaching nearly 3 million exchanges per day. Anthropic said Alibaba used the outputs to help develop its Qwen models and for other research, including reinforcement learning and model-architecture work.
Anthropic also accused Moonshot AI, the Beijing-based developer of Kimi, of routing customer requests to Claude without clearly notifying users. During one 10-day period, nearly 300,000 requests were relayed through thousands of accounts, and more than 23 million exchanges were attributed to Moonshot between May and July. Anthropic said Moonshot saved some exchanges and extracted Claude reasoning transcripts for training purposes.
DeepSeek was accused of using similar tactics, with Anthropic observing more than 12 million suspected distillation attacks over 14 days in July 2026. The company warned that some requests contained sensitive information involving individuals, multinational companies and state-affiliated actors, potentially raising privacy-law and terms-of-service concerns.
The report covers alleged misuse detected between December 2025 and August 2026 across seven areas, including cyber operations, influence campaigns, surveillance, fraud, biological misuse, conventional weapons development and AI-model distillation. The companies named did not immediately respond to CNBC’s requests for comment.
Entities: Anthropic, Alibaba, Moonshot AI, DeepSeek, Claude • Tone: analytical • Sentiment: negative • Intent: inform
11-09-2026
Y Combinator CEO Garry Tan says he would take no action against AI model distillation by Chinese competitors, despite growing pressure from Silicon Valley companies and U.S. national-security agencies. Distillation involves using the outputs of a more capable model to train a smaller one, and OpenAI and Anthropic have accused Chinese firms including DeepSeek, Moonshot AI, and MiniMax of using the technique. U.S. agencies including the NSA, CISA, and FBI have also issued a cybersecurity advisory about the practice.
Tan argues that concerns about distillation are complicated by unresolved copyright questions surrounding AI training data. He points to lawsuits involving The New York Times, Microsoft, OpenAI, and Anthropic as evidence that regulators should be cautious about restricting model development. Instead of focusing primarily on preventing distillation, he says regulators should seek a sustainable balance between open-weight models, which provide broad access and freedom, and proprietary frontier models, whose commercial viability depends on maintaining a price premium.
Tan also urges a more practical approach to AI safety. He says policymakers and the public should focus on immediate, evidence-based threats rather than hypothetical “doomsday” scenarios. He identifies cybersecurity breaches and the potential misuse of AI to develop biological weapons as more immediate concerns. Anthropic recently said it had blocked Claude access for researchers in several unspecified countries over fears that they were using the system to create bioweapons.
By contrast, Tan believes AI-driven job losses and broad economic transformation will unfold over decades. He expects workers gradually to automate routine tasks and devote more time to creative work, arguing that this gradual transition is not necessarily harmful. Despite his restrained view of long-term risks, Y Combinator continues to invest heavily in AI: 149 of the 196 startups presenting at its Demo Day were classified as machine-learning or AI companies.
Entities: Garry Tan, Y Combinator, OpenAI, Anthropic, Moonshot AI • Tone: analytical • Sentiment: neutral • Intent: inform
11-09-2026
Anthropic said it disrupted several state-sponsored surveillance operations that used its artificial intelligence models between January and July 2026. According to a report from the San Francisco-based AI company, the campaigns originated in China, Iran and West Africa and targeted diaspora and dissident communities, including Hong Kong pro-democracy activists, Tibetan and Falun Gong communities, and Iranian minorities and opponents of the Iranian government living abroad. One Iranian operation reportedly used social media accounts to identify individuals, while a contractor linked to Malian national security authorities used Anthropic’s Claude model to develop software for intelligence gathering. Anthropic said the cases illustrate how AI can substitute for a large engineering workforce in surveillance operations.
The company also said it blocked attempts to use its models to design weapons, conduct potentially dangerous biological research and create dating scams. It accused Chinese AI developers Moonshot and DeepSeek of secretly using Claude to generate answers for their own customers while extracting the data to improve competing models, a practice known as distillation. Anthropic alleged that Moonshot routed nearly 300,000 customer requests through 5,380 fraudulent accounts over ten days, potentially exposing sensitive user information and violating privacy agreements. DeepSeek allegedly used similar methods.
Anthropic said it could not determine whether researchers involved in the blocked biological-research cases intended to cause harm and withheld their identities to avoid putting them at risk. The research involved pathogens and toxins including chikungunya, highly pathogenic bird flu, smallpox- and mpox-related viruses, venoms and other toxins. Overall, the report highlights the expanding use and misuse of AI for state surveillance, cyber-enabled intelligence gathering, model theft and potentially hazardous scientific activity.
Entities: Anthropic, Claude AI model, China, Iran, West Africa • Tone: analytical • Sentiment: negative • Intent: warn
11-09-2026
The article reports that militants in Yemen used an artificial-intelligence tool developed by Anthropic in an effort to create location-guidance software for weapons. The allegation appears in a threat report released by Anthropic, the AI company, and concerns the possible use of generative AI to support the development of guided weapons. The report indicates that the militants attempted to use the system for a military application rather than an ordinary civilian or commercial purpose.
The incident is presented as part of a broader debate about the risks posed by increasingly capable artificial-intelligence systems. In particular, it raises concerns that AI tools could be used by militant groups or other harmful actors to assist with weapons development and thereby create dangers to human life. The report’s disclosure suggests that AI companies are monitoring how their systems are used and are publishing information about potentially dangerous activity.
The available excerpt does not provide further details about the militants’ identities, the specific AI tool or model involved, the technical sophistication of the location-guidance software, whether a functional weapon was produced, or whether the effort caused casualties or other damage. It also does not include Anthropic’s specific response beyond the publication of the threat report. The article’s central focus, based on the supplied text, is therefore the reported attempted misuse of an AI system for guided-weapons development and the wider public concern that advanced AI could be harmful when used by hostile or violent groups.
Entities: Pranshu Verma, The Washington Post, Anthropic, Militants in Yemen, Yemen • Tone: urgent • Sentiment: negative • Intent: inform
