Chinese tech giant Alibaba has been accused of illicitly extracting artificial intelligence models from Anthropic, the US-based AI safety company. Sources familiar with the matter have confirmed that the UK’s cyber spy agency, GCHQ, has been placed on alert amid concerns over intellectual property theft and potential national security implications.
The allegations, first reported by a financial news outlet, suggest that Alibaba’s cloud computing arm used sophisticated data scraping techniques to obtain proprietary AI training data and model architectures from Anthropic’s systems. The extraction reportedly took place over several months, exploiting vulnerabilities in Anthropic’s network security. Anthropic, known for its work on safe and ethical AI, has not publicly commented on the breach but is believed to have alerted US and UK authorities.
GCHQ’s National Cyber Security Centre (NCSC) is now monitoring the situation, with officials assessing the risk to UK infrastructure and companies that use Anthropic’s AI tools. The NCSC has issued a confidential advisory to key sectors, urging enhanced vigilance and suggesting that similar extraction attempts may be targeting other AI firms.
This incident underscores a growing geopolitical scramble for AI supremacy. China has made no secret of its ambition to lead in artificial intelligence by 2030, and state-backed entities have been accused of industrial espionage in the past. However, Alibaba, a publicly traded company with a strong global presence, has denied any wrongdoing, calling the allegations “baseless and unsubstantiated.”
The extraction method, described by cybersecurity experts as a form of “AI model inversion,” involves feeding queries to a model to reconstruct its underlying parameters and training data. While such techniques are known in academic circles, their use at an industrial scale for theft represents a new frontier in cyber-espionage. If proven, this would be one of the largest thefts of AI intellectual property to date.
The implications are profound. AI models are trained on vast datasets and require immense computational resources. Stealing a model effectively gives the perpetrator a shortcut to replicate advanced capabilities without the cost or ethical oversight. For Anthropic, which prides itself on “Constitutional AI” that aligns with human values, the theft could mean its safety measures are replicated in systems without similar guardrails.
For the UK, the alert is a stark reminder of the vulnerabilities in the AI supply chain. British companies relying on Anthropic’s models for everything from customer service to medical diagnostics could face risks if those models are compromised or weaponised by adversarial actors. GCHQ is working with the AI industry to tighten security protocols and is advising companies to audit their AI dependencies.
The story also raises questions about the effectiveness of current export controls on AI technology. While the US has restricted the sale of advanced AI chips to China, software and algorithms remain harder to police. If Alibaba or other Chinese entities can extract models remotely, the controls become porous.
As the investigation unfolds, the tech world watches with unease. This is not just a corporate dispute; it’s a tremor in the delicate balance of digital sovereignty. For Julian Vane, this is the kind of Black Mirror moment that keeps him up at night: a future where the very code that promises to help humanity is stolen and turned against us.
