In a landmark moment for AI-driven science, Anthropic announced that its Claude model discovered a previously unknown CRISPR-like enzyme system in just 21 hours. The system ran 950 autonomous agents, burned through 210 million tokens analyzing bacteriophage DNA, and uncovered a viral gene-editing mechanism that human researchers then validated in Anthropic’s brand new Bay Area wet lab. This isn’t just a new enzyme — it’s a new paradigm for how biological discovery happens.

What Happened

Anthropic CEO Dario Amodei disclosed on X that Claude identified the biological mechanism “mostly, though not entirely” on its own during a single intensive research cycle. Rather than months of manual laboratory screening, the AI autonomously screened bacteriophage genomes, pinpointed candidate CRISPR-like systems, and flagged the most promising target for physical testing.

The discovery came from Anthropic’s newly revealed wet lab operations in the Bay Area, which quietly opened earlier this spring. This is the facility’s first major scientific disclosure. Human researchers executed all physical lab experiments — DNA synthesis, cloning, and functional assays — but the computational discovery and hypothesis generation were almost entirely Claude’s work.

The screening required 210 million tokens of compute, running across 950 parallel agents. By comparison, traditional lab screening for such systems can take months and cost millions of dollars. Anthropic proved that an AI-first approach can collapse that timeline to under a day while covering far more genomic diversity.

Read the full announcement →

My Take

This is the most significant AI news of the week because it changes the story from “AI can write code” to “AI can drive the next wave of biotech.” CRISPR discovery has been largely serendipitous — Emmanuelle Charpentier and Jennifer Doudna found the first system by accident. Anthropic just showed that we can engineer that serendipity at scale.

For developers and founders, the implication is clear: the most valuable asset in biotech is no longer just lab space or PhDs — it’s the agentic infrastructure to search biological space autonomously. If you can wrap an LLM around a biological search problem, you can discover things in hours that used to take years. Every wet lab should be asking what they can automate with agent swarms.

The 950-agent architecture also matters. This is not fine-tuning a model; it’s orchestrating a distributed discovery process. The real breakthrough here is the pipeline: autonomous screening → human validation → publication. That loop is now proven.

What to Watch

  • Other AI labs rushing to replicate this approach — expect a wave of agent-driven biological discovery announcements from OpenAI, DeepMind, and smaller startups within months.
  • Anthropic’s wet lab becoming a core differentiator — if they can close the loop from AI hypothesis to physical experiment faster than anyone, they own the discovery layer of biotech.
  • The regulatory implications — CRISPR systems discovered by AI raise questions about ownership, patentability, and biosafety review that current frameworks aren’t designed for.