OpenAI AI Updates: June 17, 2026

1. OpenAI Details Deployment Simulation to Predict Model Behavior Before Release

OpenAI. OpenAI published research on Deployment Simulation, a method that uses real conversation data to predict how a model will behave once deployed, aiming to surface misaligned behaviors that traditional evaluations miss. The team analyzed roughly 1.3 million de-identified conversations spanning GPT-5 Thinking through GPT-5.4 deployments from August 2025 to March 2026, then measured the approach against two criteria: taxonomy coverage, which checks whether post-release auditing surfaces important behaviors absent from the initial taxonomy, and directional accuracy, which checks whether the simulation correctly predicts whether a known behavior’s prevalence would rise or fall after release. OpenAI says it has already used the technique during model development to identify evaluation blind spots and inform mitigations and deployment decisions. Source

2. OpenAI Brings Codex App Features to the EEA, UK, and Switzerland

OpenAI. OpenAI made several Codex app capabilities available in the EEA, UK, and Switzerland, with these features off by default in those regions. Computer Use is now available on macOS and Windows, letting Codex operate desktop apps by seeing, clicking, and typing, and the Codex Chrome extension can run browser tasks that need signed-in Chrome context, working across tabs in the background. Memories can retain useful preferences, recurring workflows, tech stacks, and repository conventions when enabled. OpenAI also opened Chronicle as an opt-in research preview for ChatGPT Pro subscribers on macOS, which helps Codex build memories from recent screen context. Source