Artificial intelligence company Anthropic has announced that new security measures designed to stop competitors from copying the capabilities of its Claude AI models have significantly reduced attempts to extract its technology, with stolen data reportedly falling by as much as 99%.
The California-based company disclosed on Monday that its latest safeguards had reduced the amount of information harvested through so-called AI distillation attacks by approximately 90 to 99 per cent.
The development comes amid growing competition in the artificial intelligence industry, particularly between American and Chinese technology companies, over access to advanced AI capabilities.
Anthropic Tightens Security Around Claude Models
Anthropic said the new protections focus on restricting access to Claude’s internal reasoning capabilities, making it more difficult for outside parties to extract information that could be used to reproduce the model’s performance.
According to the company, the amount of reasoning data extracted by parties attempting to replicate Claude is now between 10 and 100 times lower than before the safeguards were introduced.
It also said it had strengthened its systems to prevent users from repeatedly probing the models through proxy networks, which can be used to disguise the source of requests and carry out large-scale extraction attempts.
Anthropic described its detection system as highly effective, comparing its strength to protections designed to prevent Claude from assisting with the development of biological weapons.
The company said the system has a false-positive rate of less than 0.1 per cent, meaning legitimate activity is rarely incorrectly flagged by the safeguards.
AI Distillation Raises Concerns Over US-China Competition
Model distillation has become a growing concern among major AI developers, including Anthropic, OpenAI and Google.
The technique involves using the outputs of an existing AI model to help train another system to reproduce some of its capabilities. Although distillation can have legitimate applications, companies have raised concerns that competitors could use it to replicate advanced models without incurring the same development costs.
Anthropic said it had banned tens of thousands of accounts connected to suspected distillation campaigns as part of its efforts to protect Claude’s capabilities.
The company also identified Chinese AI startup Z.ai, also known as Zhipu, as an example of a firm it said had attempted to extract capabilities from Claude while developing its GLM-5.3 model.
Anthropic warned that such activities could create broader safety and national security concerns, particularly as governments and technology companies compete to develop increasingly powerful AI systems.
Z.ai had not immediately responded to a request for comment.
Anthropic Warns of Safety and National Security Risks
The company’s latest announcement highlights the growing importance of security measures as advanced AI systems become more capable and commercially valuable.
Anthropic argues that large-scale extraction of its models could allow other developers to reproduce sophisticated capabilities more quickly and cheaply, potentially bypassing safeguards built into the original systems.
Its efforts to restrict distillation come as major technology companies face increasing pressure to protect their intellectual property while preventing their AI tools from being misused.
The reported reduction in extraction attempts marks a significant development in Anthropic’s efforts to secure Claude, although the company says the threat remains an important concern amid intensifying global competition in artificial intelligence.
