OpenAI and xAI Caught Revising Model System Cards Post-Publication

ShakeelHashim · x · 2026-08-06

According to the AI safety watchdog The Midas Project, at least three different AI model system cards have been revised after their initial publication over the past two weeks.

Two revisions came from OpenAI, involving adding red-teaming data to GPT-5.6 and correcting misconfigured safety evaluations for GPT Live. Meanwhile, xAI made the most extensive changes to the Grok 4.5 system card, revising the document across the board and adding a brand new section on engineering safety. The organization tracks corporate AI safety policy changes and violations.

Original post →

More from Safety

Safety channel →