Anthropic cuts Claude token theft by 99 percent, blames Chinese AI firms
Source published ·Modelwire updated
Original coverage: The Decoder ↗·How Modelwire adds context
The development
Anthropic has substantially reduced unauthorized extraction of Claude's reasoning tokens, claiming a 90 to 99 percent reduction in large-scale distillation attacks. The company attributes the coordinated theft campaigns to Chinese AI firms Zhipu and Moonshot, and has already suspended tens of thousands of accounts involved. This escalation signals a critical vulnerability in frontier LLM economics: reasoning tokens represent high-value intellectual property that competitors can harvest at scale. The incident underscores growing tensions over model access control and raises questions about whether API-level defenses can meaningfully protect proprietary inference chains as distillation techniques mature.
Modelwire’s AI-generated summary of coverage from The Decoder.
Modelwire analysis
Analyst takeOur AI-generated reading of the wider context and the next developments to watch.
The attribution to specific Chinese firms by name is the buried lede here. Anthropic isn't just describing a generic threat category; it's publicly naming Zhipu and Moonshot, which is a deliberate escalation that carries diplomatic and commercial weight beyond the technical fix itself.
This story sits inside a week of Anthropic tightening its control perimeter from multiple directions simultaneously. The expanded TOS banning model abuse (covered October 8th across The Decoder and The Verge) established behavioral enforcement at the user level. The security research program expansion from October 7th showed Anthropic building offensive intelligence about its own vulnerabilities. Now the distillation crackdown extends that same logic to the API layer, targeting extractive use rather than abusive use. Taken together, these moves suggest a coordinated internal effort to harden Claude's value boundary before competitive pressure from cheaper models erodes the reasoning premium Anthropic charges for.
Watch whether Zhipu or Moonshot publicly dispute the attribution in the next 30 days. A non-denial would effectively confirm the scale of the operation, while a rebuttal would force Anthropic to produce evidence that could reveal how it detects distillation attempts, which is itself competitively sensitive.
This interpretation is generated from the summary above and the archive coverage cited below. Our methodology · Report an error
Coverage behind this analysis
These archive entries ground the connection in our analysis. They are ordered by source publication date, with links to our coverage and the original sources.
·The Decoder
Anthropic expands Claude access for security researchers with relaxed safety controls
Anthropic is scaling its security research program by granting more penetration testers and vulnerability researchers relaxed safety guardrails on Claude. The move reflects a strategic bet that LLM safety and real-world security are complementary rather than opposed. Prior cohorts identified over 129,000 vulnerabilities in four months, with a third rated critical or high-severity, suggesting the…
MentionsAnthropic · Claude · Zhipu · Moonshot
How this coverage is produced
Modelwire uses AI to generate summaries and context from source headlines, snippets, and selected archive coverage. Automated checks do not verify every claim, and items are not routinely reviewed by a person before publication. Zacaria Solis operates the site. Read the linked source for the full evidence and report errors through our corrections process.
Modelwire summarizes, we don’t republish. The Decoder originally reported this story as “Anthropic says it cut Claude reasoning token theft by up to 99% in fight against industrial-scale distillation”. The full content lives on the-decoder.com. If you’re a publisher and want a different summarization policy for your work, see our takedown page.