OpenAI exec envisions 25 years of science in 5. Meanwhile, Anthropic crashes software stocks.

At Cisco’s AI Summit yesterday, OpenAI’s VP for AI and Science Kevin Weil made an extraordinary claim: AI could compress scientific progress by roughly 5x.

“If we can truly accelerate science, if we can do the next 25 years of science in the next five years instead… we will be sitting there in 2030 with the technology and the science of 2050,” Weil told attendees.

Weil, formerly Chief Product Officer at Twitter and Instagram, now leads OpenAI’s newly formed AI for Science division. To back up his claim, he pointed to recent mathematical breakthroughs – specifically, a string of Erdős problems that GPT-5.2 helped solve in January, verified by Fields Medalist Terence Tao. The caveats matter: Tao characterized these as “lowest hanging fruit” – problems “amenable to simple proofs using fairly standard techniques” that happened to remain uncatalogued. GPT-5.2 scores 77% on competition-level math but just 25% on open-ended research requiring genuine insight. Still, 11 Erdős problems have been marked solved with AI involvement since Christmas – a pace that would have been unthinkable a year ago.

Anthropic, meanwhile, has recently moved markets. The company’s Claude Cowork plugins, which launched Friday with tools for sales, finance, legal and data marketing, triggered a sell-off in software stocks on Tuesday. Thomson Reuters fell 15.83% – its largest single-day drop on record. LegalZoom sank 19.68%. An ETF tracking the software industry had its worst day since April. “Why do I need to pay for software,” LPL Financial’s Thomas Shipp wrote, “if internal development of these systems now takes developers less time with AI?” Anthropic CEO Dario Amodei has warned that AI could displace “half of all entry-level white-collar jobs in the next 1-5 years.”

In the broader ecosystem, some pieces are already snapping into place. Google DeepMind announced its first “automated research lab” will open in the UK in 2026. It expects robotics capable of synthesizing and testing hundreds of materials daily. Its AI Co-Scientist platform, a multi-agent system that generates, critiques, and refines hypotheses in iterative loops, is now being piloted at academic institutions worldwide. Anthropic’s Claude for Life Sciences reportedly cut Novo Nordisk’s clinical study documentation (drafts of Clinical Study Reports (CSRs)) from over 10 weeks to 10 minutes.

But the bottlenecks are often metadata and expert review. Laboratory information systems remain fragmented. Google’s AI Co-Scientist can draft a proposal, but someone still has to know where the samples are and make sure the facts are accurate. “You still need to be a bit of an expert to get the most use out of these systems,” Stanford computer scientist Kyle Swanson told IEEE Spectrum.

Whether the underlying capabilities match the ambition remains to be seen. Both OpenAI and Anthropic are running Super Bowl ads this Sunday.  Anthropic’s mocking ChatGPT’s move to add advertising, OpenAI’s touting “builders.” Meanwhile, OpenAI has been aggressively marketing its Codex software functionality with promotional pricing after working to address infrastructure bottlenecks that have hampered performance against competitors.

While OpenAI’s science ambitions are notable, they evidently will be featured in the Super Bowl spot. OpenAI did, however, formalized its science ambitions last year with the OpenAI for Science division. The flagship product so far is Prism, a free LaTeX workspace with GPT-5.2 integrated for drafting, citations and equations. The division has also signed partnerships with the U.S. Department of Energy and Los Alamos National Laboratory.