Anthropic says AI distillation safeguards cut Claude data extraction

Traders Union
All News
Financial News
Anthropic AI distillation safeguards

Highlights

Anthropic reports that new AI distillation safeguards have reduced reasoning token extraction attempts from Claude by 10 to 100 times compared to previous levels.
The company has banned tens of thousands of accounts associated with distillation campaigns, specifically citing Chinese AI startup Z.ai’s attempts to extract Claude capabilities for its GLM-5.3 model.
Anthropic warns that unauthorized distillation raises safety and national security risks, as U.S.-China AI competition intensifies and companies protect frontier model advantages.

New protections target Claude extraction attempts

As reported by Reuters, Anthropic says safeguards that restrict access to Claude's reasoning have sharply reduced extraction attempts, with the number of reasoning tokens extracted by distillers now 10 to 100 times lower than before.

The California-headquartered company says it has made it harder to use proxy networks to repeatedly probe Claude's protected reasoning, blocking extraction attempts and complicating large-scale distillation efforts. Anthropic also says its safety system is roughly as robust as protections that prevent Claude from helping build bioweapons, with a false positive rate of less than 0.1%.

Anthropic says it has banned tens of thousands of accounts linked to distillation campaigns and cites Chinese AI startup Z.ai as an example of a company that attempted to distill capabilities from Claude while developing its GLM-5.3 model. Z.ai does not immediately respond to a request for comment.

AI competition raises security and cost concerns

Model distillation is emerging as a key concern for leading AI companies as U.S.-China competition intensifies in the sector. Companies including Anthropic, OpenAI and Google argue the practice allows rivals to reproduce advanced capabilities more quickly and at lower cost.

Anthropic warns that such efforts carry both safety and national security risks, reflecting a broader industry push to protect high-value model behavior from unauthorized replication. The issue is becoming more commercially important as AI developers invest heavily in frontier systems and seek to defend their operational advantages.

Our earlier article covered Anthropic’s $100 million training initiative aimed at developing “Frontier Deployed Engineers” who can integrate Claude into real enterprise workflows. The program combines short on-site intensives with longer workplace residencies to address the AI skills gap and accelerate practical adoption of Claude across large organizations.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.