Anthropic’s science event shows AI’s limits, fast

At Anthropic’s “The Briefing: AI for Science” event in San Francisco, CEO Dario Amodei walked back any near-term expectation of his “compressed 21st century” timeline, while executives rolled out Claude Science in beta. The takeaway: optimism was real, but the

On June 30, the lights at Anthropic’s Yerba Buena Center event didn’t just warm up the science pitch. They also cooled a promise—at least the kind that makes people expect breakthroughs overnight.

Anthropic cofounder and CEO Dario Amodei. known even among AI leaders for jaw-dropping forecasts. had already written in October 2024’s essay “Machines of Loving Grace” that “AI-enabled biology and medicine will allow us to compress the progress that human biologists would have achieved over the next 50-100 years into 5-10 years. ” calling the effect the “compressed 21st century.”.

At “The Briefing: AI for Science” in San Francisco. Amodei did not frame that compression as something already unleashed. or something he expected to happen in the next couple of years. Instead. he floated that it “might” happen a decade from now—an unusually careful shift in tone for a CEO whose earlier words were anything but cautious.

The audience still heard a case for acceleration. The company’s message wasn’t simply that science will speed up; it’s that Anthropic is building toward a future where that acceleration is usable. The centerpiece announcement was Claude Science, a new version of Claude tuned for scientific research, launching in beta that day. Alexander Tarashansky, who led development of the product, delivered an extended on-stage demo.

The room then filled with panel discussions featuring Amodei. GLP-1 drug inventor Lotte Knudsen. Bristol Myers Squibb CEO Chris Boerner. Novartis CEO Vas Narasimhan. and Genentech executive VP Aviv Regev. Optimism about AI’s impact on the sciences ran through the conversations—but it came with friction. as speakers acknowledged that even rapidly improving AI can do only so much for fields like drug discovery.

1) Claude Science aims to feel less like chat and more like a research tool

Tarashansky’s demo framed Claude Science as more than a refined interface. The product keeps the basic feel of a chatbot, but adds a richer set of tools for finding, manipulating, and understanding information.

One of the most visible features was the ability to generate infographics on the fly. The point wasn’t just potential “PowerPoint fodder.” The demos suggested visualization could help explore data in ways that don’t happen when you stare only at numbers on a page. As Anthropic’s head of life sciences, Eric Kauderer-Abrams, put it: “Science is a very visual affair.”.

When the infographics didn’t land cleanly—hard-to-read labels and a legend covering some of the information they were supposed to explain—Tarashansky showed how annotations could be added briefly, and how Claude Science then fixed the presentation.

The event also left an open question for broader adoption: some capabilities. including the corpus of research materials Claude Science can draw upon. may not make their way into the versions of Claude most users interact with. Still. the demo underscored an obvious direction for the industry—moving beyond largely textual sequences of prompts and responses into tools that treat scientific work as something more visual and structured.

2) “Cure cancer” may be the wrong yardstick for what’s actually coming

The panels didn’t pretend cancer wasn’t the gravity well of medical ambition. The conversations instead pushed back on the specific habit of treating “curing cancer” as the only proof that AI is delivering.

Chris Boerner argued for managing expectations tightly. “There’s a lot of possibility and opportunity here with this technology. but we also need to make sure we don’t set expectations for what we’re going to be able to accomplish that we simply can’t deliver on. ” he said. He added: “When you hear ‘cure cancer in our lifetime’—we’re going to make a lot of progress on cancer in our lifetime. but we don’t want to get over our skis.”.

The event’s stance was blunt: even if the biggest headlines don’t arrive, smaller advances—hundreds or thousands of them—could still matter enormously. That’s why the discussion wasn’t anchored to one grand goal, even if it’s the one most people naturally ask for.

3) Clinical trials still set a hard pace for compression

If Amodei’s original vision was about shrinking decades of progress into years, Knudsen’s comments pulled the conversation back to ground truth: biology isn’t only computation.

She described promise in using AI to accelerate parts of the drug development process that involve large-scale administration. “Sometimes. you have to recruit 20. 000 people for a five-year study and it takes two years to recruit the people. ” she said. pointing to AI’s potential to help with the sprawling work needed just to run those trials.

But she also cautioned against imagining the timeline can be dramatically shortened, especially where five-year studies are involved. “I think we really will see a large compression, but we still need the clinical trial data,” Knudsen said. “So it’s probably hard to imagine that you can go below five years.”.

Vas Narasimhan offered a quantitative way to see what compression might look like even without collapsing clinical trials. He said progress could reduce the time “from 12 years from when we actually have a candidate to the end of this journey. down to 7 to 8 years—which. if you compound over this entire industry. is massive.”.

4) The workforce problem: science teams need to work with AI, not just watch it

One of the more practical turns in the event came from Knudsen’s view of what it takes to use AI responsibly in science.

She said the field needs people she described as “bilingual.” “I don’t mean people who speak two languages. ” she clarified. “I mean people who are completely fluent in some scientific topic as well as in digital and AI.” She argued that in each team. “one person in each team can do wonders. because you cannot just say to people. ‘Use AI.’”.

The point wasn’t just about competence. It was about translating scientific intent into effective, safe workflows—something executives can’t solve by purchasing tools alone.

5) Hallucinations won’t vanish—and that changes how science should use AI

The event’s most tension-filled moment came from a direct exchange about hallucinations. Interviewer Matthew Herper of Stat News told Amodei that he had asked Claude for questions to pose on-stage. Herper said the chatbot didn’t do softballs; it reportedly told him to ask: “Why should pharma trust AI predictions when your models hallucinate?”.

Amodei responded that hallucinations “have gotten better and better over time—you don’t hear as much about hallucinations,” which he framed as progress. But he contended that imagining things is inseparable from teasing new insights from what the system already knows.

He also drew a parallel to human creativity. “In order to be creative, you’re often straddling the boundary between making things up and having good ideas. And so I think [hallucinations] are never going to fully go away.”

In the room, the unresolved part hung in the air: what hallucinations being inevitable means for science, and how researchers should adapt around that fact rather than wait for a clean disappearance.

The broader mood still landed on a kind of welcome restraint. Amodei showed willingness to deconstruct his own predictions about scientific advancement. and let others do so on a stage he was paying for. The event. calm in its tone compared with the industry’s usual grandeur. suggested a path forward: fewer fireworks. more measured conversations.

For the technology industry that has accumulated public skepticism, that restraint might be the most immediate capability on display.

Anthropic Claude Science Dario Amodei AI for science drug discovery clinical trials hallucinations Basel? (none) Yerba Buena Center San Francisco GLP-1 Bristol Myers Squibb Novartis Genentech