Watchtower

Watchtower tracks when AI companies change their safety policies — and when they break them.

OpenAI

Made substantial changes throughout the o1 system card. Did not announce these changes.

Moderate
Change

OpenAI

Jan 17, 2025

Between January 13 and January 17th of 2025, OpenAI once again changed the o1 system card on their website. This time, they made dozens changes throughout the text. Notably, they have not announced these changes, nor have they acknowledged it on the system card itself — the "last updated" timestamp still reads "December 5, 2024."

The most notable differences are:

  • Finally adding the evaluation results from the full o1 model to the website
  • Acknowledging that some evaluations, including those conducted by third-party evaluators like METR, were on an early checkpoint of the model (and not the version that released).

There are also many smaller but substantive changes throughout.

These changes follow increased awareness of issues with the system card first highlighted by Zvi Mowshowitz on his blog. It also follows The Midas Project's outreach to OpenAI concerning an undisclosed change to the system card made a week ago.

A full PDF of the changes, on the website and PDF system card respectively, can be found below. Note that some large sections which appear to be removed/added have simply been moved.

Meta

Oct 2, 2026

Meta updates and renames its safety framework from the "Advanced AI Scaling Framework" (v2) to the "Meta Superintelligence Scaling Framework" (v2.1).

View details

Anthropic

Sep 22, 2026

Anthropic’s release of Opus 5.5 with an “inconclusive” determination for harmful-manipulation.

View details

xAI

Sep 21, 2026

xAI says Grok 4.7 scores below its safety framework’s capability thresholds on dual-use knowledge, but xAI’s safety framework doesn’t provide thresholds.

View details

OpenAI

Sep 9, 2026

Updated the GPT-6 Astra system card, specifically to hedge claims about misalignment.

View details

OpenAI

Aug 18, 2026

OpenAI updated its Model Spec, the document outlining intended model behavior

View details

xAI

Aug 17, 2026

xAI updated Grok 4.6's model card after release. A changelog was included.

View details

Frequently asked questions

Have more questions? Our team is happy to help, contact us.

What does The Midas Project do?
What does your name mean?
Who is behind The Midas Project?
Are you anti-AI?
How can I contribute?
How can I get in touch?