Anthropic spent Monday positioning itself as a sufferer. By dusk, Elon Musk had flipped the script completely.
The Amazon-backed synthetic intelligence firm behind Claude publicly accused three Chinese language labs of orchestrating a large, coordinated operation to steal its mannequin’s capabilities. Hours later, Musk fired again on X with a blunt counter-accusation that despatched the story in a really completely different path.
“Anthropic is guilty of stealing training data at massive scale and has had to pay multi-billion dollar settlements for their theft. This is just a fact,” Musk posted Monday, based on Benzinga.
What Anthropic accused the Chinese language labs of doing
Anthropic’s unique declare was placing. In an in depth weblog submit Monday, the corporate mentioned DeepSeek, Moonshot AI and MiniMax carried out what it known as “industrial-scale distillation attacks” on its Claude fashions.
The three labs allegedly created greater than 24,000 fraudulent accounts and generated over 16 million interactions with Claude to extract its most superior capabilities. Anthropic mentioned the campaigns particularly focused Claude’s edge in agentic reasoning, device use, and sophisticated coding, the areas the place it instructions a significant aggressive benefit.
The operation was not clumsy. Based on CNBC, the labs used business proxy companies operating sprawling networks of fraudulent accounts to evade detection, mixing distillation visitors with regular buyer requests to mix in. When Anthropic launched a brand new mannequin throughout MiniMax’s lively marketing campaign, the lab pivoted inside 24 hours, redirecting practically half its visitors to siphon the newest system’s capabilities.
Extra AI Shares:
- Morgan Stanley units jaw-dropping Micron value goal after occasion
- Financial institution of America updates Palantir inventory forecast after non-public assembly
- Morgan Stanley drops eye-popping Broadcom value goal
Anthropic known as the risk pressing. “These campaigns are growing in intensity and sophistication,” the corporate wrote. “The window to act is narrow.”
How every Chinese language lab ran its marketing campaign towards Claude
- DeepSeek generated over 150,000 exchanges, utilizing synchronized visitors patterns and shared fee strategies to extract foundational reasoning capabilities and censorship-safe alternate options to politically delicate queries.
- Moonshot AI ran greater than 3.4 million exchanges focusing on agentic reasoning, device use, coding, and pc imaginative and prescient. It used diversified account sorts particularly to make the operation tougher to detect as coordinated.
- MiniMax was the biggest operation by far, with greater than 13 million exchanges aimed squarely at agentic coding and gear use. It redirected practically half its visitors to Claude’s latest mannequin inside 24 hours of its launch.
How Musk turned the accusations again on Anthropic
Musk’s response landed quick. He shared screenshots of X Group Notes alongside his submit, which alleged that Anthropic settled a $1.5 billion lawsuit in September 2025 for pirating greater than seven million books from shadow libraries to coach Claude.
The notes additionally flagged that the corporate at the moment faces a separate $3 billion lawsuit from music publishers over alleged copyright infringement involving greater than 20,000 songs.
Picture by Bloomberg on Getty Pictures
“They built their models on stolen creative work, then complain when others extract their outputs,” one Group Observe learn, based on Benzinga. Anthropic has not publicly confirmed both settlement and didn’t instantly reply to requests for remark.
Musk additionally known as the corporate “super smug, sanctimonious and hypocritical” in a follow-up submit, a pointed jab at Anthropic’s well-cultivated safety-first model id.
Why the hypocrisy argument is gaining traction
The criticism Musk leveled isn’t completely new. OpenAI, Google, Meta and most main synthetic intelligence firms face ongoing lawsuits from authors, information publishers, and musicians who argue their copyrighted work was used with out permission to coach AI methods.
Based on TechCrunch, Anthropic’s accusations additionally arrive at a fragile second, because the Trump administration just lately loosened export controls on superior Nvidia chips to China, a coverage Anthropic has actively lobbied towards.
The authorized traces round distillation stay genuinely blurry. Courts are nonetheless wrestling with whether or not AI mannequin outputs qualify for copyright safety in any respect. The U.S. Copyright Workplace affirmed in January 2025 that copyright requires human authorship, which suggests Anthropic could face actual challenges if it pursues authorized motion towards the Chinese language labs.
Musk’s personal xAI enterprise has confronted comparable questions on how Grok was skilled. The talk round information provenance touches each main lab within the trade, and that broad publicity is precisely what made his counter-punch so efficient on Monday.
What this feud means for the AI trade
The stakes right here transcend a social media spat. Anthropic’s distillation claims, if confirmed, characterize a major risk to the economics of constructing premium AI fashions. If rivals can extract years of analysis and a whole bunch of thousands and thousands in coaching prices by means of coordinated scraping, the aggressive moat round frontier fashions shrinks significantly.
Anthropic is valued at $380 billion following its newest funding spherical, with Amazon as its anchor investor. xAI raised $6 billion final 12 months. Each firms are racing towards the identical vacation spot, and the battle over information is now as fierce because the battle over compute.
Musk’s submit landed the sharper punch on Monday. Anthropic got here in because the accuser and left the information cycle defending itself. That shift in narrative is the true story, and it’s one all the AI trade might be watching rigorously as copyright battles and information wars proceed to escalate.
Associated: Elon Musk has robust phrases on new Apple-Google AI transfer
