Anthropic says new defenses blunt efforts to copy Claude
Anthropic, the AI developer behind Claude, reports that new measures appear to cut the reasoning data attackers can collect by roughly 90 to 99 percent. The company also named China's Z.ai as one of the copiers.
By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .
This story is no longer being updated. It reflects reporting up to .
What happened
Anthropic, the AI developer behind Claude, said new methods have cut the reasoning data attackers harvest from its model. Early evidence suggests the reduction is roughly 90 to 99 percent. The company also named China's Z.ai as one of the copiers.
Why it matters
The measures target large-scale model copying, proxy networks and suspicious accounts, amid growing concern over unauthorised AI training.
Timeline
- Oct 12, 2026: Anthropic details new methods aimed at distillation attacks on Claude.
- Oct 13, 12:22 AM ET: Anthropic says the safeguards have reduced reasoning data extracted through suspected copying by 90 to 99 percent.
- Oct 13, 2:39 AM ET: Anthropic names China's Z.ai among the companies copying Claude.
More from Technology & AI
- Eight states sue OpenAI, alleging misleading AI safety claims
- OpenAI announces solution to Navier-Stokes math problem
- Datadog bets AI agents can handle testing of AI-written code
- Anthropic and OpenAI sketch wealth-sharing plans before IPOs
- Mamdani administration proposes AI customer service rules
- Florida engineers study how AI behaves in high-risk situations