Thinking Machines Proposes Phased Open Weights Safety Path

On August 1, Thinking Machines Lab published "A Safe Path to Open Weights," outlining its safety strategy for releasing open weight models. The article argues that indiscriminately releasing weights is unsafe, but confining powerful models to a few labs is equally undesirable, advocating for a balanced, phased expansion of access. Co-founder Mira Murati shared the article, and research scientist John Schulman expressed support, stating the company loves open weights and plans to continue releasing models and fine-tuning tools, but is not absolutist because abuse risks are real. On the same day, the company announced it is expanding its safety team and hiring. Commenter keirp1 praised the gradual openness approach for its transparency and logical coherence, criticizing other AI labs for poor handling of safety and open source issues.

Confirmed

• Thinking Machines published an article proposing a phased open weights safety path.
• Mira Murati shared the article; John Schulman endorsed the stance, emphasizing a non-absolutist approach.
• The company announced the expansion of its safety team and open hiring.

Unconfirmed

• Specific timelines, access conditions, and model details for the phased rollout have not been disclosed.

Why it matters

This strategy attempts to offer a third path in the open source versus safety debate and could influence industry practices and regulation around releasing open weight models.

2026-08-01 ~ 2026-08-02 · 5 related posts

• [source] Thinking Machines Lab Proposes Staged Open-Weights Release for Safe AI — thinkymachines · 2026-08-01
• [source] Thinking Machines Expands Safety Team, Hiring Now — thinkymachines · 2026-08-01
• Mira Murati's Think Machine Outlines Safe Path for Open-Weight AI — clarejtbirch · 2026-08-01
• The Open Weights vs. Safety Debate: Staged Access as the Middle Path — keirp1 · 2026-08-01
• [source] John Schulman Outlines Safe Open-Weight Path, Advocates Staged Access — ZhongRuiqi · 2026-08-02