What changed
OpenAI described Deployment Simulation, a privacy-preserving method that regenerates responses to de-identified historical conversation prefixes to estimate how often undesirable behavior may appear after launch.
Why it matters
Traditional red teams find severe edge cases but say little about incidence. Simulation connects pre-release testing to the distribution of actual use.
What to do
Replay representative, privacy-reviewed historical contexts against candidate versions, pre-register predictions, and compare them with post-release behavior.
Caveat
The method cannot reliably estimate failures rarer than roughly one in 200,000 messages in the reported setup, and realistic tool resampling remains difficult.