Profluent builds the ‘ChatGPT for CRISPR’ and open-sources it

For years, discovering a new gene editor meant combing through hot springs, peat bogs, animal poo, and even yogurt containers, hoping nature had evolved the right molecular scissors.

AI startup Profluent just showed there is a much faster way. Instead of scouring the planet for novel microbes, they used generative protein language models to design functional gene editors completely from scratch.

Even better: the company just open-sourced its flagship AI-designed editor, OpenCRISPR-1, for broad ethical use in research and commercial therapies.

Learning the language of CRISPR

Traditional protein engineering relies heavily on copying what already exists in nature or tweaking amino acids one by one through directed evolution.

The math behind that approach is brutal. A typical Cas9 protein spans over 1,000 amino acids, creating a design space with more possible sequences than there are atoms in the observable universe. Change a single amino acid in the wrong spot, and the entire system falls apart.

To bypass that random evolutionary cycle, Profluent built the CRISPR-Cas Atlas, an unprecedented dataset spanning 5.1 million CRISPR proteins that expanded known natural diversity by 2.7-fold.

They then trained large protein language models on roughly 240,000 Cas9 variants. The model generated 4 million new sequences, effectively expanding the diversity of known CRISPR families by 4.8-fold in just a few computer runs.

400 mutations away, yet fully functional

Generating sequences on a screen is one thing, but making them cut DNA inside living human cells is an entirely different hurdle.

Profluent took its top candidate, dubbed OpenCRISPR-1, and tested it directly in human cells. Despite having more than 400 mutations separating it from standard SpCas9, the AI-designed enzyme matched or exceeded the real-world cutting activity and target specificity of its natural counterpart.

It is compatible with standard guide RNA workflows, meaning researchers can drop it into existing pipelines without reinventing their entire lab setup.

By releasing the sequence publicly, Profluent wants to establish a community baseline for AI-generated biologics rather than keeping the molecule locked behind proprietary gates.

Why it matters

CRISPR has already won a Nobel Prize and earned FDA approval, but the entire space remains tangled in brutal patent battles and licensing hurdles.

We think OpenCRISPR-1 represents a massive geopolitical and economic shift for biotech. By designing an enzyme that performs identically to natural Cas9 while carrying over 400 distinct sequence differences, Profluent has essentially engineered a clean escape hatch around existing patent thickets.

For academic labs and clinical startups, having an open-source, high-performance editor lowers the barrier to exploring rare disease treatments dramatically.

That said, clearing the benchtop is only step one. Any novel synthetic protein still faces intense scrutiny over off-target cuts, delivery packaging, and immunogenicity once injected into living patients.

Even with those clinical hurdles ahead, one thing is clear: biology is shifting rapidly from an observational science to an intentional, prompt-driven engineering discipline.