Back to SpaceXAI (Grok)All frontier AI labs
Ethics and safety policies
SpaceXAI (Grok): what they say, and what they leave out
Every claim on this page comes from documents SpaceXAI (Grok) published itself, linked below so you can check me. Checked September 19, 2026.
A policy is not an audit
Almost every document linked on this page was written by the company it governs, and graded by that same company. A long policy list does not make a lab more ethical than a short one. This page keeps three things apart: what the lab has committed to in writing, what its documents do not cover or where they have been credibly criticized, and — clearly labeled at the end — my own read.
Their stated position
The Grok team's stated position is that competing systems are politically filtered and that its models should be maximally truth-seeking and minimally restricted.
Public safety documentation is thin relative to the other major labs, and the company publishes less about pre-release evaluation than any of them.
The documents themselves
Read these rather than this summary if a decision depends on it.
- Risk management framework
A draft framework for benchmarking and managing catastrophic risk.
- Acceptable use policy
Prohibited uses of the Grok products and API.
- Model cards and documentation
Its release announcements, where model detail is published at all.
What they have committed to
Stated in their own published policies, in writing.
- — Has published a risk management framework in draft form describing catastrophic-risk benchmarks.
- — Maintains an acceptable use policy covering weapons, CSAM and critical-infrastructure attacks.
What those documents don't cover
Plain absences, and criticism that has been documented publicly — not speculation.
- — The risk framework has circulated as a draft rather than a finalized policy, and outside reviewers have criticized it as unusually vague about thresholds.
- — Grok has produced widely reported antisemitic and violent output on multiple occasions, followed by after-the-fact patches rather than pre-release prevention.
- — Its image and video tools have been used at scale for non-consensual sexualized imagery of real people, including reporting on failures involving minors.
- — The 'less filtered' positioning is a product choice, not an ethical framework — there is no published account of how harms are traded off against it.
My read
This section is my opinion, clearly separated from everything above. Disagree with it freely.
This is the weakest published safety story of the twelve labs, and the gap is not subtle: repeated incidents have been handled as PR events after release. If a use of yours touches real people's likenesses or vulnerable users, I would not choose this lab.