Google denies training Bard on ChatGPT data

Did Google get caught copying OpenAI’s homework?

A explosive report from The Information alleges that Google trained its Bard chatbot using data from OpenAI’s ChatGPT, scraped from a prompt-sharing site called ShareGPT.

Google quickly denied the accusation. Spokesperson Chris Pappas told The Verge that Bard is not trained on any data from ShareGPT or ChatGPT. However, when asked whether the company had ever used ChatGPT data in the past, Google refused to answer, merely reiterating its blanket denial.

The defection of an engineer

The most dramatic detail in the report involves a key Google AI researcher named Jacob Devlin.

According to the report, Devlin warned Google executives not to use ChatGPT data to train Bard. His argument was twofold: it directly violates OpenAI’s terms of service, and it would cause Bard’s answers to look far too similar to its rival.

Devlin reportedly resigned over the issue and immediately jumped ship to join OpenAI. One source noted that Google did stop using the scraped data after Devlin’s warnings, and may have scrapped that portion of the training entirely.

DeepMind called to the rescue

Beyond the data controversy, the report reveals just how panicked Google is over Bard’s underwhelming debut.

Google is reportedly forcing its DeepMind unit to join forces with the Google Brain team on an initiative codenamed Gemini, specifically built to take on OpenAI.

This is a massive internal shift. DeepMind has famously spent years trying to gain more operational independence from Google. Now, the parent company is pulling DeepMind right back into the core product trenches to salvage its AI reputation.

Why it matters

Using output from a frontier model to train smaller models—often called distillation—is common practice in open source, but it is a massive legal and reputational minefield for a tech titan like Google.

If Bard were indeed running on recycled ChatGPT answers, users would simply be getting a degraded carbon copy rather than true technological competition. For an industry desperate for genuine alternatives to OpenAI, that would be a huge letdown.

Google’s aggressive denial stops short of addressing what happened earlier in development. We think the real story here is the desperation behind the scenes. Pushing DeepMind and Google Brain together under Gemini shows Google is now throwing every resource it has at the problem. Whether an uncomfortable internal merger can outrun a fast-moving OpenAI, however, remains to be seen.