Anthropic’s AI discovery now has a provenance problem

Anthropic said Claude helped discover a new biological system. A researcher says his team found it first.

Mario Rodríguez Mestre has studied the same enzymes since 2022 and says his team shared unpublished findings with Claude while using it for code, manuscripts, and research tasks. Anthropic says Claude was not trained on user transcripts and its biology team had no access to them.

The unresolved part matters. Anthropic described its AI as taking “the first steps of a biological discovery on its own,” yet it has not disclosed the precise prompts used in the experiment.

That makes independent verification difficult, especially when Mestre says Anthropic’s system analyzed the same 2022 dataset his team had already examined and discussed with Claude.

You can almost hear the laptops closing. Mestre says he is shutting down his Claude projects, while other scientists are now questioning what happens to the unpublished work they type into AI tools.

Researchers need clearer provenance before “AI discovery” becomes a credible scientific claim.

What will scientists still be willing to share?

Meta wants AI power users testing Muse first

Meta is letting Muse recruit its own early testers.

Users can ask the AI to register them for early access with a simple prompt, giving Meta a pool of people who already use AI products heavily. That is more interesting than a standard random beta because these users are likely to compare Muse directly with competing agents.

The features they will test are substantial. Meta has teased a video-chat avatar, more shopping connections, Mac controls that can complete tasks on your computer, and Muse integration with its AI glasses.

I think the computer access matters most. Once an assistant can move through apps on your laptop, clicking and completing tasks, the quality of its actions matters as much as the quality of its answers.

Meta also gets something valuable here: feedback from users who know what other AI agents can already do.

The next test is simple. Will people trust Muse with the mouse?

AI can spot aging inside a cell nucleus

A microscope image may contain more information about aging than researchers can see.

Scientists in Barcelona built ChromAgeNet, an AI model that examines 3D images of blood stem cell nuclei and identifies subtle changes in how DNA is organized. It classified cells as young or aged with 77% accuracy.

The useful part is the setup. The researchers used DAPI, a cheap and common DNA stain, which means labs could potentially run this analysis without expensive new equipment or complex molecular tests.

Under the microscope, many of these aging differences are too subtle for the human eye. ChromAgeNet picked up patterns involving chromatin organization and could also detect when drug-treated old cells shifted toward a younger-looking nuclear state.

That does not mean the cells were rejuvenated. The study did not establish that.

Still, I think the screening angle is worth watching. A low-cost imaging test could let researchers run large numbers of compounds through aging experiments before spending more time and money on deeper biological testing.

How useful will that 77% become at scale?

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.