Headline: White House accuses China-linked firms of “industrial-scale” AI model theft — Anthropic names suspects; US vows a response
The White House says foreign actors—primarily linked to China—are running “industrial-scale” campaigns to extract capabilities from American AI models, and those campaigns could undercut US firms and weaken safety protections.
Key points
- The White House Office of Science and Technology Policy (OSTP) announced the allegation, saying foreign entities are using unauthorized “model distillation” to copy features from US AI systems.
- Michael J. Kratsios, assistant to the president for the science office, said the government has “information” that these groups rely on tens of thousands of proxy accounts and jailbreaking techniques to evade detection and probe models.
- OSTP warned that models built from these surreptitious distillation campaigns may not match the originals on full performance, but can mimic them on select benchmarks at far lower cost—and might lack important safety or neutrality controls.
- The allegations follow a February claim from Anthropic. The developer of the Claude model accused three Chinese firms—DeepSeek, Moonshot, and MiniMax—of conducting distillation attacks that produced roughly 16 million exchanges with Anthropic’s models using about 24,000 fraudulent accounts.
- Anthropic said the attacks targeted capabilities such as coding, agentic reasoning, data analysis, grading tasks, and computer vision.
- The OSTP said the administration will work with US companies to share information on large attacks and coordinate stronger defenses, and will “explore measures to hold foreign actors accountable,” though it did not list specific penalties or enforcement steps.
- The announcement comes amid intensifying US–China competition over frontier AI, which US officials frame as central to national security and economic power.
Why this matters for crypto audiences
- Many crypto projects rely on proprietary AI for trading bots, risk modeling, smart contract auditing, on-chain analytics, or user-facing services. Unauthorized replication could let lower-cost competitors offer comparable capabilities or enable adversaries to bypass safety controls.
- AI-model theft raises intellectual property and operational risk for tokenized AI services and startups that monetize models through usage-based (token) pricing—copied models can disrupt pricing and market trust.
- Weakening of safety controls in copied models could also increase fraud or manipulation risks in crypto ecosystems that depend on AI-driven decisioning.
What’s next
The administration says it will collaborate with industry to detect and defend against these campaigns and consider accountability measures. But details on enforcement remain sparse, and the episode underscores growing geopolitical stakes around access to and protection of advanced AI capabilities.
Read more AI-generated news on: undefined/news
