Anthropic, the American artificial intelligence lab founded by former OpenAI researchers, has formally accused Chinese tech giant Alibaba of illicitly extracting proprietary model weights and training data. The accusation, filed in a confidential submission to UK and US regulators, has triggered a classified review by British intelligence agencies concerned about industrial espionage and national security.
According to sources familiar with the matter, Anthropic’s internal forensic analysis traced suspicious API calls from Alibaba’s cloud infrastructure to a series of extraction tools that systematically queried Claude’s response patterns. The tools appear to have reconstructed large portions of the model’s neural architecture, effectively reverse-engineering weeks of work in just hours.
The implications extend far beyond corporate rivalry. If proven, this would be the largest theft of intellectual property in the AI sector, involving a model trained on billions of dollars worth of compute. But what worries Whitehall officials more is the potential for transferred capabilities. Advanced AI models, when replicated, could be fine-tuned for surveillance, disinformation, or even autonomous weapons systems.
Alibaba has vehemently denied the allegations, calling them ‘baseless and competitive slander’. The company pointed to its own open-source AI models as evidence of its commitment to transparency. Yet the UK’s National Cyber Security Centre (NCSC) is quietly liaising with US counterparts to verify the claims, while the National Security Council has been briefed on risks to the UK’s AI supply chain.
This is not a lone incident. In recent months, multiple AI firms have reported anomalous probing from state-linked entities. The Anthropic-Alibaba case, however, marks the first time a Western company has openly named a Chinese tech giant in an espionage context. The fallout could reshape global AI governance, accelerating calls for a ‘Digital Sovereignty’ framework that protects intellectual property while maintaining open research.
For the average user, this may feel distant, but it is deeply personal. Every interaction with AI systems relies on trust in the safety and integrity of their underlying models. If models can be stolen and weaponised, the ‘User Experience’ of society itself is compromised. Governments must now decide whether to enforce stricter export controls, akin to those on nuclear materials, or risk a fragmentation of the internet into rival AI ecosystems.
The UK, caught between its ‘Global Britain’ ambitions and its Five Eyes alliances, finds itself in a delicate position. It cannot afford to alienate China, a critical trading partner and potential AI collaborator. Yet it cannot ignore evidence of systematic theft. The next weeks will test whether British diplomacy can balance innovation with security, or whether we are witnessing the opening salvo of an AI cold war.
As Julian Vane, I have long warned of the Black Mirror consequences of unregulated AI. This is not a dystopian fiction. It is happening now, in data centres and committee rooms, where the battle for our digital future is being waged. The Anthropic-Alibaba affair is a watershed moment. How we respond will define not just the AI industry, but the very nature of trust in the digital age.
