# Labs Ship Reduced-Safeguard Model Tiers to Vetted Customers

Rather than choosing between withholding a dangerous capability and releasing it openly, frontier labs increasingly split releases into a restricted-behavior public model and a lighter-safeguard variant gated to vetted institutions, so expect recurring two-tier launches, growing contested questions about who qualifies for the privileged tier, and the vetting list itself becoming a lever of industrial and national policy.

- Conviction: 44 / 100 (weakening)
- 7-day move: +8
- Horizon: Emerging (watchlist)
- Tracking since: 2026-09-06T00:00:00.000Z
- Last updated: 2026-09-14T14:04:09.672Z
- Canonical: https://polylog.news/ai/trends/lab-safeguard-tiering-by-customer
- Publisher: Polylog
- Affected regions: United States

## Recent score history

- 2026-09-13: 46
- 2026-09-14: 44

## Recent evidence

- [confirms] Meta Claims Coding Leadership for Muse Spark 1.3, With Its Best Numbers From a Restricted Variant (2026-09-12): Meta's headline Muse Spark 1.3 numbers come from a restricted variant that broad developer populations cannot reach, while a weaker configuration is what ships publicly. The two-tier release pattern the thesis tracks is recurring here on capability access rather than safeguards, extending the mechanism to performance gating.
- [confirms] Anthropic's Fable 5.1 Doubles Its Own Science Terminal Score, and Its Unrestricted Twin Beats It on Coding (2026-09-10): Anthropic's Mythos 5.1 — the same model as the public Fable 5.1 but with looser cyber and biology safeguards — scores 60.9 percent on Terminal-Bench 4.0 against 55.8 percent for the released version. A measured 5-point capability premium attached to the reduced-safeguard tier turns access to that tier into a quantifiable commercial advantage, which is what makes the vetting list a policy lever rather than a formality.

2 more evidence entries, the full score history, the conviction-driver timeline, and affected assets are for subscribers: https://polylog.news/pricing