Timnit Gebru warns AI APIs risk reproducibility, control
According to a post by Timnit Gebru referencing Zack Korman on X, reliance on AI lab-controlled APIs threatens reproducibility and governance as providers can change behavior at any time.
SourceAnalysis
On November 14 2026 Timnit Gebru shared a post quoting Zack Korman that highlighted how much software now depends on AI APIs that providers can alter without notice. The post noted the lack of reproducibility guarantees and the contrast with concerns over sleeper agents in open models.
Korman pointed out that AI labs hold stronger control through API access than through any backdoor in released weights because they can modify behavior at any time. This dependency reshapes developer workflows by pushing teams toward self-hosted or open-weight alternatives to maintain consistent outputs and audit trails across projects.
timnitGebru (@dair-community.social/bsky.social)
@timnitGebruAuthor: The View from Somewhere Mastodon @timnitGebru@dair-community.