Watchtower
Watchtower tracks when AI companies change their safety policies — and when they break them.
Reported to have abandoned former promise to dedicate 20% of compute resources to advanced AI alignment.
OpenAI
May 21, 2024
In July 2023, OpenAI announced the Superalignment team, which was intended to research how superintelligent AI systems could be aligned to human values. As part of this announcement, OpenAI also promised that they would give 20% of their computing resources to AI alignment research.
However, within a year, multiple leading members of the team resigned, accusing OpenAI of putting their bottom line over safety and responsibility. One of the accusations included that they never provided the promised computing resources to the team.
Since then, OpenAI is widely reported to have fully abandoned the promise they made in July 2023. We cannot confirm or deny this, as they have never spoken about it.
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).
Anthropic
Sep 22, 2026
Anthropic’s release of Opus 5.5 with an “inconclusive” determination for harmful-manipulation.
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.
OpenAI
Sep 9, 2026
Updated the GPT-6 Astra system card, specifically to hedge claims about misalignment.
OpenAI
Aug 18, 2026
OpenAI updated its Model Spec, the document outlining intended model behavior
xAI
Aug 17, 2026
xAI updated Grok 4.6's model card after release. A changelog was included.
Frequently asked questions
Have more questions? Our team is happy to help, contact us.