Anthropic says its new biology lab has already found something big. TechCrunch AI reports that the company’s Bay Area wet lab has found a previously unknown enzyme system with “properties reminiscent of CRISPR,” the gene-editing tool that changed modern biology. The announcement came just a week after Anthropic confirmed the lab exists. According to CEO Dario Amodei, Claude did most of the work.
🧬 The Quick Version
- What: A new enzyme system hidden in the DNA of bacteriophages (viruses that infect bacteria).
- What it does: Anthropic says it can “perform operations like cutting, copying, and pasting DNA,” much like CRISPR.
- Who found it: “Mostly, though not entirely, by Claude,” Amodei wrote on X.
- How fast: 21 hours of focused compute, using about 950 agents and 210 million tokens.
- Who did the lab work: Human scientists. Claude doesn’t touch the equipment.
- The catch: Nobody outside the company has confirmed it yet, and a Stanford team had already found a system “in some ways similar.”
Why This Matters
The discovery itself will need outside validation. But the process is the real story. A lab that opened only this spring (Anthropic won’t say exactly when) produced a candidate discovery in months. The AI’s share of the search took less than a day.
It’s a clear example of an AI lab using agent swarms for real scientific discovery instead of chat or coding. About 950 agents working through huge volumes of genomic data is a very different job from answering questions. If the result holds up, it’s strong evidence that a large multi-agent setup can do the pattern-hunting that usually eats weeks of a researcher’s time.
Amodei is also open about the limits. He credits earlier work and points to the Stanford team’s similar find. That honesty helps. It frames Claude as a fast accelerator of existing science, not a lone genius.
⚠️ The Safety Tension
The timing is hard to ignore. AI CEOs, Amodei among them, have recently said publicly that frontier models are getting capable enough, and risky enough, that the industry needs to slow down and build better safety testing. Earlier this month, a couple of Anthropic employees said publicly that AI could pose an existential risk.
Amodei has also named bioterrorism as one of his biggest fears about AI. So an Anthropic biology lab raises obvious questions. The company’s answer is a set of tight limits:
- Work stays at biosafety levels BSL-1 and BSL-2 (the lowest risk tiers).
- The lab doesn’t handle pathogens that can infect humans.
- Every physical experiment is done by a human scientist.
What stands out here is that last point. Claude suggests what to test, and people decide what actually gets done. That line matters more than the enzyme right now.
Anthropic Isn’t Alone
AI-driven biology is already a crowded field:
- Stanford researchers just published work combining LLMs and CRISPR.
- UC San Francisco teams have used AI to design enzymes from scratch.
- Google DeepMind’s AlphaFold has been transforming protein structure prediction since 2020.
What’s different is that a frontier model developer now runs its own wet lab. That lets it test its models’ predictions directly against real experiments. It’s a tight loop that most AI companies don’t have.
What to Watch Next
- Peer validation. Other researchers will judge how new and how useful this enzyme system really is. Expect some pushback on the “CRISPR-like” framing.
- More lab announcements. Anthropic clearly sees this as a flagship use case. Amodei still says AI could “cure most diseases in 5-10 years.”
- Automation creep. Amodei hasn’t ruled out Claude running the lab itself someday. “Eventually it may even be possible for Claude itself to safely perform the experiments by autonomously controlling lab equipment, with appropriate safeguards in place, but we aren’t doing that today,” he said.
- Regulatory attention. AI plus biology plus fast agents is exactly the mix policymakers worry about. Expect more questions about oversight.
For practitioners, it’s worth noticing the pattern: hundreds of parallel agents working on a narrow, data-heavy search problem, with humans checking the results. That setup works well beyond biology.
The next few months will show whether this enzyme becomes a real research tool or just a well-timed headline. Either way, AI labs have clearly moved into physical science. Full details are available in the original TechCrunch AI report.