OpenAI ramps Model Spec transparency + safety bug bounty

Daily Tech Intelligence (Last 48h — 2026-03-25)

OpenAI shipped a dense cluster of policy + safety moves (Model Spec transparency deep-dive, new Safety Bug Bounty focused on agentic abuse/prompt injection/data exfil, and an open-source teen-safety policy pack for gpt-oss-safeguard). On product, ChatGPT shopping/product discovery is getting more visual and is expanding ACP integrations with major retailers.

Also on the OpenAI YouTube channel: a podcast episode explaining the Model Spec’s chain-of-command, how it’s updated, and why transparency matters; plus a short Notion clip showing Codex enabling a “voice input” feature ported from mobile to web/desktop rapidly.