A major international row has erupted after Alibaba, the Chinese e-commerce giant, was accused of stealing proprietary artificial intelligence technology from Anthropic, a leading AI safety company backed by British investors. The allegations, which surfaced in a leaked internal memo, claim that Alibaba’s cloud computing division illicitly obtained and replicated core components of Anthropic’s language model architecture, including its constitutional AI alignment system.
Anthropic, founded by former OpenAI researchers Dario Amodei and Daniela Amodei, has positioned itself as a champion of ethical AI development, with a focus on ensuring that advanced systems remain aligned with human values. The company’s approach, which relies on a ‘constitution’ of principles to guide model behaviour, has attracted significant investment from UK-based venture capital firms and the British government’s AI Safety Institute.
The memo, reviewed by this publication, details an internal investigation within Alibaba that allegedly uncovered a deliberate effort to reverse-engineer Anthropic’s technology. Engineers within Alibaba’s DAMO Academy (the company’s research arm) are said to have used publicly available API access to extract and replicate key features of Anthropic’s Claude model, including its ‘harmlessness’ fine-tuning and constitutional training data.
Alibaba has vehemently denied the accusations, calling them ‘baseless and defamatory’. In a statement, a spokesperson said: ‘Alibaba has always respected intellectual property rights and operates in full compliance with international laws. Our AI technology is the result of years of independent research and development.’ The company has threatened legal action against those spreading the allegations.
However, sources close to Anthropic suggest that the US-based firm has already shared evidence with regulators in the United States and the United Kingdom. The UK’s Competition and Markets Authority (CMA) is said to be ‘monitoring the situation closely’, given the national security implications of AI technology being diverted without authorisation.
The accusation comes at a tense time for global AI governance. The United States has been pushing for stricter controls on the export of advanced AI technology to China, while the UK has positioned itself as a neutral arbiter through initiatives like the AI Safety Summit. If proven, this breach could severely damage trust between Western AI developers and Chinese tech firms, potentially leading to a fragmentation of the global AI research community.
Dr. Eleanor Shaw, a digital ethics researcher at the University of Cambridge, commented on the broader implications: ‘This isn’t just about one company. It’s about whether the global AI ecosystem can sustain open collaboration while national security concerns lead to more closed development. The UK backs Anthropic because it wants a safe, transparent approach to AI. If that technology is stolen, it undermines the entire model.’
The technical aspects of the alleged theft are particularly concerning. Anthropic’s constitutional AI methodology is one of the few proven techniques for aligning models with complex human values without extensive human feedback. If Alibaba has indeed managed to replicate this, it could accelerate their own AI development by years, potentially enabling them to deploy high-stakes systems without the same safety precautions.
Alibaba’s cloud division has been aggressively expanding its AI offerings, competing with global leaders like Amazon and Microsoft. In recent months, the company launched a series of large language models aimed at enterprise customers, some of which bear striking similarities to Anthropic’s Claude in their ability to refuse harmful requests. While imitation is common in AI, the specificity of these parallels has raised eyebrows.
As the story develops, all eyes will be on how both governments respond. The UK has invested significant political capital in being a global leader in AI safety, and the protection of British-backed innovation lies at the heart of that strategy. For now, the tech world watches, waiting to see if this is an isolated incident or a harbinger of a new kind of digital Cold War.
