US AI company Anthropic announced that it will put 5% of its shares into a public welfare fund, to be used for the public interest and to address the impact of AI on the labor market.
On October 12, Claude developer Anthropic released a follow-up article naming Zhipu and Moonshot, stating that it had banned tens of thousands of accounts related to distillation attacks and rejected trillions of tokens of paid traffic. It estimates that the new defenses have reduced the volume of extracted inference data by approximately 90% to 99%.
On October 12, the author of the WeChat public account CroWall posted again, saying that their Claude account was banned in the early hours of the 11th, that they suspected they had been subjected to entrapment, and explaining that removing sensitive words was done to avoid the app being taken down and the account being banned.
On October 10, the author of the WeChat public account CroWall wrote that when using Claude to create a promotional image for a music player, the playlist name in the Simplified Chinese version showed the words “Taiwan independence,” while the Traditional Chinese version read “Mandarin pop.” The author proactively took the app down and asked Claude to remove the sensitive words.
On September 14, the Chinese Ministry of Foreign Affairs said that the US AI giant's spreading of threat narratives, engaging in confrontation and malicious competition is not in the interest of any party. Earlier, the CEO of US AI giant Anthropic called for slowing AI development, receiving responses from many quarters; BBC analysis believes that coordination between China and the US on AI is almost impossible, but both sides intend to include AI in the summit agenda.
In mid-September 2026, Anthropic CEO Amodei called on frontier companies to jointly slow the pace of AI development in order to buy time for establishing safety standards, winning support from OpenAI CEO Altman, Musk, and others.
On September 13, Anthropic released its September threat intelligence report, stating that operators linked to local Chinese police and state security agencies had used Claude to assist in monitoring petitioners, rights defenders, and overseas dissidents, with the "Teacher Li Is Not Your Teacher" account being one of the monitoring targets.
Anthropic co-founder Amodei published a long article arguing that China's lead in artificial intelligence poses a serious threat to the United States and the world, and backing Washington's restrictions on selling powerful AI chips to China; Chinese state media Global Times said his claims of containing China are an unrealizable delusion, and China's Foreign Ministry said it has noted the relevant remarks.
Anthropic released a report stating that the company discovered and banned a batch of accounts associated with local Chinese public security and state security departments. These accounts used Claude to collect information, query surveillance databases, and generate daily reports. "Teacher Li Is Not Your Teacher" was explicitly listed as one of the targets for stability maintenance and public opinion monitoring.
On September 12, 2026, Anthropic CEO Dario Amodei posted on social media, calling for a slowdown in frontier AI development, stronger international cooperation, and the introduction of third‑party assessment bodies to track safety practices. Subsequently, OpenAI CEO Sam Altman expressed agreement with the call and revealed that the originally planned listing has been postponed to 2027 due to safety concerns. The statements from the two CEOs stem from former Anthropic and OpenAI employee Jacob Coxon warning on X and in an NPR interview that the two companies are "gambling with our lives" and expressing worry that rapid AI progress could pose a lethal threat to humanity within a decade. Amodei emphasized that AI remains beneficial to humanity, but must be built in the right way.
On September 11, 2026, the US AI giant Anthropic released a long-form investigative report detailing how Chinese AI companies such as Moonshot AI and DeepSeek, while distilling own models from US models like Claude, allegedly sent real internal prompts and sensitive data of users from the People's Liberation Army, police, and state-owned enterprises to US AI platforms, sparking strong concern over model distillation security and data leakage.
On September 10, Anthropic released a threat intelligence report stating that Chinese AI companies, including DeepSeek and Moonshot, used third-party API proxy services to bypass regional restrictions and call Claude. The report also noted that these relay services may obtain model access through methods such as fake accounts and stolen API keys.
On September 10, 2026, Anthropic released a report stating that operators linked to Chinese local police and state security systems used Claude to carry out "stability maintenance" surveillance, with "Teacher Li Is Not Your Teacher" explicitly listed as a monitoring target; Anthropic said it had banned the related accounts.
Anthropic's report disclosed that a Chinese local state security agency turned its workflow into an internal AI usage manual and required collecting intelligence on overseas activities such as the assembly points, routes, and endpoints of democratic marches in Vancouver.
An Anthropic report revealed that an affiliated account linked to China's public security and state security systems asked Claude to generate control recommendations for 10 Chinese citizens, including intercepting petitioners, conducting interviews, and monitoring their movements and communications.