Verified, then logged.
Sun 22 Nov 2026 · 00:21 KST / 15:21 UTC한국어 | ENSearch /

04 / Policy · Policy

Carnegie researchers propose checks on non-public AI

ZDNET Korea reports a policy proposal addressing the limits of oversight focused on commercial AI services.

Sample vials, a well plate and a pipette sit on a biosafety lab bench before a monitor showing a protein structure model, with a researcher blurred in the background, seen from behind. AI-generated image

Key points

  1. 01ZDNET Korea's 15 Nov 2026 coverage of Carnegie's independent verification proposal
  2. 02Pre-release and internal research models as the proposed subjects of checks in Trust but Verify
  3. 03A policy recommendation with no reported implementation procedure or timetable

ZDNET Korea, a Korean technology news outlet, published coverage of the Carnegie Endowment for International Peace's proposals for oversight of non-public AI models on 15 Nov 2026. The researchers propose an independent verification framework; the coverage describes a policy recommendation.

The coverage draws on Trust but Verify, a report from Carnegie, a US think tank.

Models covered. According to ZDNET Korea, the proposed independent verification covers models awaiting formal release and models used for developers' own research and development. The outlet names AI developers OpenAI and Anthropic as companies developing such models.

Scope of oversight. ZDNET Korea reports the researchers' finding that the most technically advanced AI models operate mainly within companies. The researchers argue that oversight focused on commercial services has difficulty identifying potential risks.

Research background. ZDNET Korea describes the report as a follow-up that develops earlier policy research. According to the coverage, that earlier work helped lay the foundation for California's Frontier AI Transparency Act, SB 53, a state law.

The coverage gives no follow-up procedure or timetable for introducing the proposed verification framework.