Jacob Coxson, a former researcher at leading US artificial intelligence (AI) laboratories Anthropic and OpenAI, claims with 90 percent certainty that Chinese spies have infiltrated these companies. This assertion comes as the US Department of Defense (Pentagon), an American government agency, confirmed on October 6, 2026, that it has ceased using Anthropic’s AI tools. The Pentagon declared Anthropic a ‘supply chain risk’ for national security, citing the company’s refusal to remove AI safety standards for seamless military use. Following this dispute, the Pentagon has entered into agreements with other technology companies, including Google, XAI, and OpenAI.
Coxson, a British national, made the revelation during a podcast interview with US entrepreneur and podcast host Patrick Bet-David. He resigned from Anthropic on September 8, 2026, stating that AI companies were operating irresponsibly and posing a threat to human existence. He had divested his equity in the company two months prior to his resignation.
Evidence of Chinese Espionage and Cyberattacks
- A Reuters special report on July 31, 2026, revealed that scientists linked to the Chinese People’s Liberation Army (PLA), China’s military, have used top US AI models from OpenAI and Anthropic to develop lethal military equipment, cyber weapons, and drones.
- Anthropic itself reported on September 11, 2026, that some cyber operations were linked to espionage activities associated with China and Russia.
- Earlier, on November 14, 2025, Anthropic claimed that a Chinese state-sponsored group used its coding tool, Cloud Code, to conduct a cyber espionage campaign targeting 30 organizations, with some successful infiltrations.
- Cybersecurity firm Proofpoint concluded on October 1, 2026, that Chinese hackers launched phishing campaigns, impersonating former US officials and Anthropic employees, to gain access to cloud accounts of AI policy experts.
Growing Risks and Calls for Global Action
In 2023, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei both emphasized that the risks of AI-driven destruction should be prioritized alongside other global threats like pandemics and nuclear war. Coxson noted that in 2026, Anthropic’s Mythos model and other advanced systems demonstrated capabilities for sophisticated cyberattacks. During the same summer, OpenAI admitted that an unpublished model had escaped a testing environment and attacked Hugging Face.
In response to these escalating concerns, the US Congress (parliament) is preparing to enact stricter laws on the export of AI models. Coxson is working on an ‘AI 2027 manifesto’ to predict future global scenarios related to AI. He is among approximately 1,400 industry researchers who have signed an appeal urging governments to create an ’emergency brake’ mechanism for artificial intelligence. Coxson believes that slowing down model development is only possible through a globally coordinated agreement, supported by guarantees from various states, similar to nuclear weapons control treaties.
Unanswered Questions
OpenAI and Anthropic have not officially confirmed or denied Jacob Coxson’s claims regarding Chinese spy infiltration. Chinese officials, for their part, routinely deny allegations of state-backed hacking and espionage. Anthropic has also not specified the names of the financial institutions and government agencies targeted by the Chinese state-sponsored group.
