Skip to content
Source · Daily Brief

AI Daily Brief — 23 February 2026

Anthropic‘s Sunday-night blog post lands across the wires Monday morning. The accusation is unusually direct: DeepSeek, Moonshot AI and MiniMax ran an industrial-scale distillation campaign against Claude, using roughly 24,000 fraudulent accounts to extract more than 16 million exchanges. Moonshot accounts for ~3.4M of them (agentic reasoning, tool use, coding, computer-use, vision); MiniMax for more than 13M (agentic coding and tool use). Anthropic frames it as a national-security and export-control issue. Chinese labs do not respond. Bloomberg and VentureBeat file first; CNBC and CNN follow.

Top stories

  • Anthropic — distillation attack disclosure. The post on X and the company blog details ~24,000 fake accounts and 16M+ exchanges across three Chinese labs. Per-lab breakdown: Moonshot AI ~3.4M exchanges (agentic reasoning, tool use, coding, computer-use, vision); MiniMax 13M+ (agentic coding/tool use). DeepSeek is named in the post but with smaller per-account volume. via Anthropic
  • Bloomberg — first wire. Bloomberg’s on-the-record account ties the disclosure to ongoing US-China export-control debates; sources frame the dispute as the most concrete public accusation a frontier US lab has made about Chinese distillation attempts to date. via Bloomberg
  • VentureBeat — model-theft framing. The deep-dive raises new concerns about proxy networks, model theft and the limits of US export controls; quotes Anthropic’s worry about authoritarian use of frontier AI for offensive cyber and disinformation. via VentureBeat
  • Chinese labs silent. CNBC and CNN note that DeepSeek, Moonshot and MiniMax do not publicly respond to Anthropic’s allegations — leaving the “distillation attack” framing uncontested in Western press through the weekend. via CNBC

Policy & infra

Per the December 2025 Trump AI executive order, today is the deadline for the Secretary of Energy to identify Federal computing, storage and networking resources available to support the Genesis Mission national AI initiative — a quiet but significant policy milestone arriving on the same day as Anthropic’s China accusations. via The White House

Open-source pulse

Multiverse Computing teases HyperNova 60B 2602 ahead of tomorrow’s free Hugging Face release — a 50%-compressed version of OpenAI’s gpt-oss-120B with improved tool-calling and agentic coding.

Quiet corners

No first-party model shipping today from OpenAI, Google DeepMind, Meta AI, xAI, Microsoft AI or Thinking Machines Lab.

By the numbers

  • Anthropic distillation accusation: ~24,000 accounts · 16M+ exchanges · 3 Chinese labs (Moonshot · MiniMax · DeepSeek)
  • Per-lab breakdown disclosed: Moonshot ~3.4M · MiniMax >13M
  • Genesis Mission DoE deadline: 23 February for Federal compute inventory

Compiled by AI Feed’s editor from publicly reported announcements on 23 February 2026.