Report: Google AI engineer Jacob Devlin quit to join OpenAI in January after complaining that Bard was being trained with ChatGPT data; Google denies the charge
Google's Bard hasn't exactly had an impressive debut — and The Information is reporting that the company is so interested …
Context & Ripple Effects
Google had only recently introduced Bard as an experimental ChatGPT rival, while related reporting described an internal race to organize Google Brain and DeepMind around Gemini. This report puts training-data provenance alongside that competitive pressure.
The allegation also fits a coverage arc in which Bard was released amid reported staff reservations about its reliability, later followed by a broader rollout. Google’s denial is therefore central: the report documents a disputed claim, not an established account of Bard’s training.
First-order effects
- Devlin’s departure gives OpenAI a senior AI researcher while leaving Google to manage the internal and external fallout from a disputed claim about Bard’s development.
- Google must defend the integrity of Bard’s training process as it continues the product’s rollout; the company explicitly denies the allegation.
Second-order effects
- A public dispute over whether a rival’s output informed training raises the value of documented data lineage and internal review for frontier-model teams, particularly when employees raise concerns.
- The reported pressure to compete with GPT-4—including the Gemini collaboration between Google Brain and DeepMind—can make governance disputes more consequential, because they become part of the talent and credibility contest between labs.
Third-order effects
- If similar disputes recur, competitive advantage in generative AI will depend not only on model capability but on whether labs can demonstrate defensible training-data and governance practices.
- The episode points to frontier AI development becoming more institutionalized: rapid launches may increasingly be evaluated through employee safeguards, auditability, and public trust as well as product performance.
The trend: This is one data point in the institutionalization of frontier AI competition, where speed to market is increasingly constrained by provenance, safety, and talent-governance questions.