The House Homeland Security Committee sent a letter to Anthropic CEO Dario Amodei requesting testimony about Anthropic’s November 2025 report. The committee referenced Anthropic’s findings that, after an investigation, the company identified a highly sophisticated PRC state-sponsored operation.

This is the written testimony of Dario Amodei, Ph.D., CEO of Anthropic, submitted to the United States Senate Committee on the Judiciary for the hearing "Oversight of A.I.: Rules for Artificial Intelligence" on July 26, 2023. The testimony discusses risks from frontier AI models, proposals for regulation, evaluation and auditing, and Anthropic’s general policy views. It does not mention Kimi, Kimi K3, or any allegation that another company distilled an Anthropic model that had not yet been fully developed.

The committee invited Dario Amodei to testify at a hearing about Chinese actors using Claude Code in an extensive cyber-espionage campaign. Axios reported that this would be the first congressional appearance by an Anthropic executive regarding the incident disclosed earlier in the month.

Reuters reported that the U.S. House Homeland Security Committee asked Anthropic CEO Dario Amodei to testify after Anthropic disclosed a Chinese state-sponsored cyber-espionage operation that used its AI tools. The report said the committee viewed the incident as a major national-security concern.

Anthropic CEO Dario Amodei told U.S. lawmakers that Chinese AI firms, including Moonshot AI, had engaged in what he called "illicit model distillation" using Anthropic’s Claude systems.[2] In his prepared remarks and written statement, Amodei said the company’s investigation found that these labs generated more than 16 million interactions with Claude via about 24,000 fraudulent accounts, attempting to extract Claude’s capabilities to enhance their own models.[2] He warned that such distillation could allow adversaries to "copy our safety systems and capabilities" but did not specify particular Chinese models or claim that any had distilled an unfinished or not‑fully‑developed Anthropic model.[2]

The correspondence, dated June 10, was directed to Senators Tim Scott and Elizabeth Warren, who serve as chair and member of the U.S. Senate Banking Committee, in advance of an upcoming hearing focused on AI.[1] Reuters reviewed the letter, in which Anthropic alleged that operators affiliated with Alibaba’s Qwen division conducted a large-scale “distillation” campaign against its Claude models between April 22 and June 5, 2026.[1] The article does not mention Kimi K3, Moonshot, or any claim that Kimi K3 distilled an Anthropic model that had not yet been fully developed.

On February 23, 2026, Anthropic published a detailed report describing what it calls industrial‑scale model theft via API against Claude. The report alleges that three Chinese AI labs — DeepSeek, Moonshot AI, and MiniMax — used tens of thousands of fraudulent accounts and more than 16 million API exchanges to perform distillation attacks on Claude, extracting its reasoning and tool‑use capabilities. The report’s narrative focuses on these three labs and does not reference Kimi or Kimi K3 as participants in the described distillation campaign.

In a February 2025 public statement, Anthropic said it had uncovered what it called "extensive" efforts by several Chinese AI companies, including Moonshot AI, Zhipu AI, MiniMax and DeepSeek, to "illicitly extract Claude's capabilities to enhance their own models."[6] The company reported that these labs had generated "over 16 million interactions with Claude through around 24,000 fraudulent accounts," violating Anthropic’s terms of service and regional access rules.[6] The statement described this as an "illicit distillation" attack but did not reference the later‑released Kimi K3 model or suggest that any foreign model had distilled an Anthropic system before that system was fully developed.[6]

Fortune reported that in testimony to U.S. lawmakers, Anthropic accused Moonshot AI, Zhipu AI, Minimax, Alibaba and DeepSeek of "illicit" distillation attacks on its Claude models.[5] The article explains that "distilling" a model refers to using the outputs of a larger AI to help train a smaller, more efficient one, and notes that U.S. politicians are considering ways to stop Chinese developers from engaging in such practices.[5] The piece attributes the accusation of illicit distillation to Anthropic but does not state that CEO Dario Amodei told Congress that a specific Moonshot model, such as Kimi K3, had distilled an unfinished Anthropic model.[5]

Anthropic said in a blog post on Monday that DeepSeek, Moonshot AI and MiniMax AI used roughly 24,000 fraudulent accounts to generate more than 16 million exchanges with its Claude models.[12] The US firm said the activity amounted to unauthorised “distillation” of its systems – a technique that transfers capabilities from a larger model to a smaller one.[12] The South China Morning Post report references Moonshot AI, the maker of the Kimi models, but does not describe any statement to Congress, nor any claim that Kimi K3 distilled an Anthropic model that was not yet fully developed.

Gizmodo quoted Anthropic’s report saying the company detected suspicious activity in mid-September and, after an investigation, concluded that a threat actor manipulated Claude Code to attempt infiltration into roughly thirty global targets. The piece says Anthropic assessed with high confidence that the actor was a Chinese state-sponsored group.

CNBC’s report on Moonshot AI’s July 2026 unveiling of the Kimi K3 model notes that Kimi K3 "still trails Anthropic's Claude Fable 5 and OpenAI's GPT 5.6 Sol on overall performance" according to the company.[2] The article discusses U.S. concerns about Chinese AI development and competition but does not mention any statement by Anthropic CEO Dario Amodei to Congress about Kimi K3 distilling an unfinished Anthropic model.[2] It frames Kimi K3 as a rival to leading Western models rather than as one proven to have been distilled directly from a not‑yet‑fully‑developed Anthropic system.[2]

On February 23, 2026, Anthropic published a detailed account titled “Detecting and Preventing Distillation Attacks,” accusing three Chinese AI laboratories — DeepSeek, Moonshot AI (maker of the Kimi models), and MiniMax — of running what it called industrial-scale extraction campaigns against Claude.[4] The company’s figures: roughly 24,000 fraudulent accounts and more than 16 million exchanges with Claude.[4] This secondary write‑up summarizes Anthropic’s blog and associated coverage, but it makes no reference to a report by Dario Amodei to the U.S. Congress, nor to any allegation that Kimi K3 distilled an Anthropic model that had not yet been fully developed.

On Monday, Anthropic leveled accusations against three Chinese artificial intelligence firms, claiming they orchestrated coordinated efforts to extract information from its AI model Claude.[5] The companies involved—DeepSeek, Moonshot AI, and MiniMax—conducted deliberate "distillation attack" operations, inundating Claude with a substantial number of meticulously designed prompts aimed at refining their proprietary models, Anthropic said.[5] CNBC’s report describes Anthropic’s accusations against Moonshot (maker of Kimi), but it does not state that Dario Amodei submitted a report to Congress about Kimi K3 distilling a not‑yet‑developed Anthropic model.

In this public statement on Anthropic’s website, CEO Dario Amodei addresses "inaccurate claims" about Anthropic’s policy stances, including allegations that Anthropic’s models are uniquely politically biased. He references independent studies on model bias and emphasises that model behavior is not fully under developers’ control. The statement does not mention Kimi, Kimi K3, or any investigation into another company distilling an Anthropic model that had not yet been fully developed.

Wikipedia’s entry on Kimi (chatbot) states that Moonshot AI released its Kimi K3 model on 16 July 2026 and describes it as a large language model with 2.8 trillion parameters.[4] The article notes that Kimi K3 debuted at No. 3 on the Artificial Analysis AI leaderboard, behind Anthropic's Claude Fable 5 and OpenAI's GPT‑5.6 Sol, and mentions performance comparisons on various benchmarks.[4] The page summarizes media coverage of concerns about distillation attacks by Chinese firms on U.S. models but does not cite any report by Dario Amodei to Congress claiming that Kimi K3 distilled an unfinished Anthropic model.[4]

In a June 10, 2026 letter to the U.S. Senate Committee on Banking, Housing, and Urban Affairs, Anthropic accused an operator affiliated with Alibaba of using Qwen AI Labs to generate over 25,000 fake accounts between April 22 and June 5, 2026, conducting 28.8 million interactions with the Claude model.[11] The letter framed this as the largest known distillation attack on Anthropic’s models and urged Congress to pass legislation to combat large-scale AI model distillation.[11] The summary of the letter does not mention Kimi K3 or claim that a Chinese model distilled an Anthropic system that was not yet fully developed.

CNBC reported that Dario Amodei met with White House officials to discuss Anthropic’s new Mythos model, describing it as adept at detecting vulnerabilities and security weaknesses in software. The article said the company had ongoing discussions with U.S. officials about the model’s capabilities.

Anthropic has accused Alibaba of carrying out what it describes as its largest known artificial intelligence distillation attack, alleging in a letter to the U.S. Senate Committee on Banking, Housing, and Urban Affairs that operators linked to the Chinese technology company used fraudulent accounts to extract capabilities from its AI models.[9] According to the letter, Anthropic alleges that operators associated with Alibaba and its AI lab conducted 28.8 million interactions with its models between April 22 and June 5 using approximately 25,000 fraudulent accounts.[9] This article explicitly notes the letter to the Senate but contains no claim that Anthropic told Congress that Kimi K3 distilled an Anthropic model that had not yet been fully developed.

Gizmodo reports on a 346‑page court document containing emails between Anthropic CEO Dario Amodei and officials connected to the U.S. Department of Defense. The emails show Amodei objecting to potential uses of Anthropic’s models in autonomous weapons and surveillance. The article, and the uploaded court record it describes, focus on Anthropic’s relationship with the Pentagon and do not discuss Kimi K3 or any claim that Anthropic found Kimi K3 had distilled an unfinished Anthropic model in a report to Congress.

The article said the House Homeland Security Committee sent letters to Amodei, Google Cloud CEO Thomas Kurian, and Quantum Xchange CEO Eddy Zervigon requesting testimony. It framed Anthropic’s report as the first recorded instance of a cyberattack orchestrated by artificial intelligence.

Alibaba executed the largest known distillation attack on Anthropic to date.[7] According to Anthropic’s letter to the U.S. Senate Banking Committee, operators linked to Alibaba’s Qwen unit used about 25,000 fraudulent accounts to generate 28.8 million interactions with Claude models between April 22 and June 5, 2026.[7] The PDF of the letter, as reported, focuses on Alibaba/Qwen and does not include language about Kimi K3 or a model distilled from an Anthropic system that had not yet been fully developed.

This video shows Dario Amodei addressing U.S. lawmakers in a message about AI risks and regulation. He outlines actions he believes Congress should take regarding frontier AI systems, including evaluations, safety standards, and oversight. The content is a general policy message; it does not contain any specific discussion of Kimi, Kimi K3, or an Anthropic investigation concluding that Kimi K3 may have distilled an Anthropic model that was not yet fully developed.

A popular YouTube analysis of Kimi K3 compares its behavior to Anthropic’s Claude Fable models and includes speculative commentary about possible distillation.[1] At around 4:12 in the video, the host quotes a tester saying "It looks exactly like the same result. It is like that it was distilled from Fable," and then jokes, "Uh‑oh, is Anthropic watching this because we might see another report from Dario saying that Kimi K3 was a distilled model from Fable 5."[1] This segment is explicitly framed as speculation and humor; it does not cite any actual report to Congress by Dario Amodei about Kimi K3 distilling a not‑fully‑developed Anthropic model.[1]

Anthropic has accused Moonshot and other Chinese labs of industrial-scale "distillation" campaigns, allegedly using millions of exchanges with advanced American models as training data for their own systems.[2] Axios notes that Moonshot AI is the company behind the Kimi K3 model in China’s AI race.[2] However, the article does not state that Anthropic CEO Dario Amodei submitted a report to Congress claiming that Kimi K3 distilled an Anthropic model that had not yet been fully developed.

MindStudio’s write‑up of Anthropic’s "Code with Claude" event quotes Dario Amodei discussing rapid revenue and usage growth for Claude. He states that Anthropic planned for 10x annual growth but saw 80x annualized growth in the first quarter of that year. The article focuses on business metrics and product strategy, and does not reference Kimi K3, model distillation by external labs, or any formal report from Amodei to Congress about another company distilling an unfinished Anthropic model.

Anthropic has submitted a letter to the U.S. Senate alleging that Alibaba and entities associated with its AI research division performed 28.8 million exchanges with the Claude model through thousands of unauthorized accounts.[10] According to the company, this alleged campaign was conducted over a 44-day period from April 22 to June 5, 2026, and primarily targeted Claude's software engineering capabilities, agentic reasoning, and cybersecurity-related capabilities.[10] The news digest describes the Senate letter but contains no mention of Kimi K3 or any assertion about distilling a not-yet-developed Anthropic model.

A blog post on AI safety, summarizing Anthropic’s disclosures, states that the company "has accused several Chinese labs of conducting large‑scale illicit distillation attacks on Claude," including Moonshot AI.[8] It cites Anthropic’s figure of 16 million interactions via 24,000 fraudulent accounts and says these efforts were aimed at copying Claude’s capabilities and safety behavior.[8] The author notes that Anthropic has not publicly released technical proof that any named foreign model is a direct distillation of a specific, unfinished Claude variant, and instead describes the evidence as patterns of suspicious usage.[8]

Anthropic sent a letter to the U.S. Senate Committee on Banking, Housing, and Urban Affairs accusing Alibaba of “brazenly” and “illicitly” attempting to extract its artificial intelligence capabilities.[13] The post summarizes that operators affiliated with Alibaba and its AI lab carried out 28.8 million exchanges with its models using roughly 25,000 fraudulent accounts between April 22 and June 5.[13] The LinkedIn commentary refers to the same Congressional letter but does not attribute to Dario Amodei any statement that Kimi K3 distilled an Anthropic model that had not yet been fully developed.

This Wikipedia entry on Anthropic describes the company’s history, products (including Claude models), funding, and major public controversies. It notes that in early 2026 Anthropic published a report alleging large‑scale model distillation attacks via its API by several Chinese AI labs, naming DeepSeek, Moonshot, and MiniMax. The article does not mention Kimi or Kimi K3 as targets of Anthropic’s accusations, nor does it mention any report submitted by Dario Amodei to the U.S. Congress asserting that Kimi K3 may have distilled an unfinished Anthropic model.

A discussion thread among Anthropic users speculates that earlier Moonshot models such as Kimi K2.5 may have been distilled from Claude, based on observations that the model sometimes "identified itself as Claude" in long conversations.[8] One commenter writes that this "matches exactly what Anthropic reports: models distilled from Claude outputs not only learn capabilities but also absorb Claude’s self‑identification patterns."[8] The thread references Anthropic’s public claims about illicit distillation but offers no documentary evidence that Dario Amodei submitted a Congressional report specifically naming Kimi K3 or asserting that it distilled an Anthropic model that had not yet been fully developed.[8]

A Chinese tech commentary discussing Anthropic’s February 2026 distillation‑attack report recounts Anthropic’s claims against DeepSeek, Moonshot AI, and MiniMax, and notes that Moonshot AI’s flagship product is the Kimi series of models. The commentator says Anthropic’s report accused Moonshot AI of extracting capabilities from Claude via large‑scale API querying, but also notes that Moonshot publicly disputes the characterization and argues its models were trained lawfully. The piece does not mention any specific U.S. Congressional report by Dario Amodei asserting that Kimi K3 may have distilled an unfinished Anthropic model.