AI models need more data about biology, and OpenAI is paying to create it

There is something profoundly poetic—and deeply weird—about teaching AI to understand biology using the corpses of companies. We tend to think of scientific progress as a linear march of successes: the breakthrough, the discovery, the miracle drug. But the reality is messier. Behind every blockbuster therapy lies a mountain of failures—trials that flopped, molecules that misbehaved, manufacturing processes that fell apart. And crucially, that knowledge is usually locked away as trade secrets, invisible to the world. Bankruptcy, it turns out, is the great unlocker. When a biotech company dies, its secrets don't die with it. They go to auction. The strange genius of Teslo's proposal is that it treats negative data as gold. AI models are ravenous for information, and they don't care whether a result was a triumph or a catastrophe—they just need to see patterns. A failed clinical trial contains a wealth of information about what doesn't work, which is often more valuable than what does. By resurrecting the data of dead companies, we're not just feeding AI more examples; we're giving it the hidden history of an entire industry's failures. That's the kind of knowledge no living company would ever volunteer. This raises a delicious irony: the very institutions that spend billions protecting their secrets are, in their moment of death, handing those secrets to the machines that will replace them. OpenAI's involvement signals that this isn't just a quirky academic idea—it's a viable strategy for building the next generation of medical AI. The economics are bizarre too. A bankrupt company's data, once valued at millions, can be had for pennies on the dollar at auction. It's like finding a first-edition manuscript in a dumpster. But beneath the clever arbitrage lies a deeper philosophical shift. We're entering an era where the value of information is no longer tied to its commercial success. The AI doesn't care if the data comes from a Nobel Prize winner or a startup that burned through its funding in eighteen months. It just wants patterns. And in that sense, the failed biotech companies of the world are not failures at all—they're unwitting contributors to a grand, collective intelligence. Their