Anthropic Accuses Alibaba of Massive 'Distillation Attack' to Extract AI Capabilities
Anthropic alleges Chinese firms used thousands of accounts to extract Claude outputs, while new open models from DeepSeek and Liquid AI emerge.
The story
A significant development this week saw Anthropic accuse Alibaba and its affiliated AI lab of a large-scale "distillation attack" on its Claude models. Anthropic claims that nearly 25,000 fraudulent accounts were used to execute over 28 million prompts within a 45-day period, aiming to extract and replicate Claude's capabilities for model training.
This accusation highlights escalating concerns over intellectual property and competitive practices within the rapidly evolving AI landscape. The alleged campaign underscores the intense pressure on AI labs to safeguard their proprietary models amidst a global race for advanced AI. The incident raises questions about the technical and legal defenses available to protect AI innovations from sophisticated extraction efforts.
Who moved
Anthropic
What Changed: Anthropic accused Alibaba and affiliated Chinese AI companies of a large-scale “distillation attack” on its Claude models.
Consequence: This highlights growing concerns around intellectual property theft and competitive tactics in the AI space.
DeepSeek
What Changed: DeepSeek released V4-Pro, a 1.6-trillion-parameter open-weight mixture-of-experts model.
Consequence: This provides developers with a powerful, freely downloadable model for complex coding and reasoning tasks.
Liquid AI
What Changed: Liquid AI launched LFM2.5-230M, a 230-million-parameter model capable of running on low-power devices like a Raspberry Pi.
Consequence: This demonstrates a push towards highly efficient, on-device AI for broader accessibility and diverse applications.
Lenovo
What Changed: Lenovo expanded its Hybrid AI Advantage with a new portfolio of AI inferencing and agentic AI innovations.
Consequence: This aims to help organizations deploy AI more securely, reduce token costs, and scale AI across various environments.
M-Files
What Changed: M-Files introduced new AI capabilities to automate document-centric processes.
Consequence: This enhances metadata enrichment, conversational access, intelligent search, and workflow automation for enterprise clients.
New Relic
What Changed: New Relic announced the next evolution of its platform to power agentic AI-first businesses.
Consequence: This focuses on providing tools for organizations to confidently embed AI into core workflows.
New models
DeepSeek V4-Pro
Lab: DeepSeek
What: A 1.6-trillion-parameter open-weight mixture-of-experts model with a 1M-token context window.
Use: Designed for frontier-grade coding and complex reasoning, available for developers to download and use freely.
LFM2.5-230M
Lab: Liquid AI
What: A 230-million-parameter model that runs efficiently on low-power devices.
Use: Enables on-device AI applications, decoding at high speeds on devices like a Raspberry Pi, and can translate plain-English commands into tool calls for robotics.
Market signals
French insurtech Alan raised €480M led by Prosus, achieving a €5.5B valuation.
Implication: This significant funding round highlights continued investor confidence in vertical AI applications, particularly in the healthcare and insurance sectors, with a focus on AI-assisted prevention models.
OpenAI's internal data indicates 97.9% of its employees now use Codex agents, with non-developer usage increasing over a hundredfold since late 2025.
Implication: This suggests strong internal adoption and utility of AI agents within a major AI lab, pointing towards the increasing integration of AI tools across various enterprise functions, not just development.
Builders saying
Yann LeCun, Former Meta Chief AI Scientist. His team's work on world models has achieved a 48x acceleration in planning, indicating rapid progress in AI systems that understand physics and spatial reasoning.
What we'll be watching
- CINECA Open Hackathon (Virtual Event, June 30 - July 8, 2026)
- AWS Summit - DC (Washington DC, June 30 - July 1, 2026)
- GITEX Europe 2026 (June 30 - July 1, 2026)
- Davos Tech Summit (Davos, Switzerland, July 1-4, 2026)
- Festival der Zukunft (July 2-5, 2026)
- OpenAI DevDay (San Francisco, September 29, 2026), with applications due by July 10, 2026.
Reporting + analyst voices: grounded via Google Search at publish time.