Back to MetaAll frontier AI labs
Ethics and safety policies
Meta: what they say, and what they leave out
Every claim on this page comes from documents Meta 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
Meta's public position is that openness is itself a safety strategy: releasing weights lets outsiders find flaws that a closed lab would not.
It publishes a Frontier AI Framework describing capabilities it says it would not release, plus acceptable-use policies for its open models.
The documents themselves
Read these rather than this summary if a decision depends on it.
- Frontier AI Framework
The outcomes it says it will not enable, and how it assesses them before release.
- Llama acceptable use policy
What downstream users of the open weights are permitted to do.
- Responsible use guide
Its recommended safeguards for developers building on Llama.
- Meta AI research responsibility
Its stated approach, tooling and open safety models such as Llama Guard.
What they have committed to
Stated in their own published policies, in writing.
- — Publishes model cards and safety tooling (Llama Guard, Prompt Guard) alongside open releases.
- — States it will withhold releases it assesses as enabling catastrophic cyber or bio outcomes.
- — Keeps weights downloadable, which allows independent evaluation nobody needs its permission to run.
What those documents don't cover
Plain absences, and criticism that has been documented publicly — not speculation.
- — Once weights are public, the acceptable-use policy is unenforceable in practice — safety fine-tuning can be stripped by anyone with a GPU.
- — Meta calls its releases 'open source' while the license restricts use by the largest competitors and imposes conditions; the Open Source Initiative disputes the label.
- — The company dissolved its responsible-innovation team in 2022 and has repeatedly restructured integrity work.
- — Training-data provenance is contested in active litigation, including allegations about the use of pirated book archives.
My read
This section is my opinion, clearly separated from everything above. Disagree with it freely.
Open weights really do enable outside scrutiny, and that is worth something no closed lab provides. It also means every stated safeguard is advisory the moment the file is downloaded — both things are true at once.