White House AI and Crypto Czar David Sacks fired a sharp public broadside at Anthropic on Tuesday, 28 July 2026, accusing the AI safety company of applying one rule to itself and another to the rest of the world when it comes to intellectual property and model training data.

In a post on X, Sacks wrote that Anthropic 'maintains that it is entitled to train for free on all the world's output, even if the author objects,' while simultaneously treating a competitor's use of Anthropic's own output — even after paying for it — as 'IP theft.' He called the contradiction 'breathtaking hypocrisy.'

Context

The charge lands in the middle of a long-running legal and ethical battle over how generative AI companies build their training datasets. Since at least 2023, content creators, publishers, and coders have alleged that major AI labs ingested copyrighted material without licences or compensation to train large language models. Courts in the United States have been asked to decide whether such ingestion qualifies as fair use — a question that remains unresolved.

Anthropic, founded by former OpenAI employees and known for its safety-focused research, has itself faced lawsuits from authors and music publishers over alleged unauthorised use of copyrighted text in training its Claude family of models. The company has publicly argued that training on publicly available data is consistent with fair use doctrine.

Policy Backdrop

Sacks's role as the Trump administration's AI and Crypto Czar gives his criticism unusual institutional weight. His office has been tasked with shaping federal policy on artificial intelligence, including questions of data governance, liability, and competitiveness. A senior White House official publicly calling out a leading AI lab for inconsistency on IP is a signal that the administration is watching how companies frame their legal positions.

The broader pattern Sacks is pointing to — AI firms claiming broad rights to ingest third-party content while aggressively protecting their own outputs — has become a central tension in both US and EU AI policy debates. Model weights, system prompts, and synthetic outputs are increasingly treated as proprietary assets even as the inputs used to create them remain contested territory.

Stakeholders and Impact

The asymmetry Sacks describes affects two distinct groups. Content creators — writers, journalists, artists, and coders — have argued for years that their work is being used without consent or payment to build commercial AI products. At the same time, AI developers who build on top of or fine-tune existing models find themselves accused of IP infringement the moment they touch a competitor's outputs, even when those outputs were accessed through paid channels.

For India, where a large and growing community of developers, publishers, and creative professionals are both consumers and potential training-data sources for global AI systems, the outcome of these debates will shape the legal and commercial landscape for domestic AI development. Indian policymakers have been watching US copyright litigation closely as they draft their own AI governance framework.

What's Next

Rulings in several pending AI copyright cases in US federal courts are expected to clarify the boundaries of fair use for model training. Any executive action or congressional legislation on mandatory data-licensing requirements — an area Sacks's office could directly influence — would reshape the economics of AI development globally. Anthropic has not publicly responded to Sacks's post as of the time of publication.

The episode underscores a growing pressure point: as AI companies mature and accumulate their own proprietary assets, the 'open internet as training data' argument they relied on in their early years becomes harder to sustain without acknowledging the rights of original creators.