As of November 30, 2025, we have not yet reached the prediction’s evaluation window (“around late 2026” / roughly three years after the October 30, 2023 AI executive order), so it is too early to determine if it came true.

The October 30, 2023 Executive Order defines covered models and infrastructure using specific, fixed compute/parameter-like thresholds (e.g., dual‑use foundation models trained with >10^26 FLOPs, lower thresholds for primarily biological sequence data, and similar capacity thresholds for large compute clusters). These are precisely the sort of model‑size/compute‑based standards Friedberg was referring to as potentially becoming obsolete.(bidenwhitehouse.archives.gov)

Subsequent U.S. policy work through 2024–2025 (e.g., Commerce rulemakings and the emerging AI Diffusion Framework) still uses very similar absolute compute thresholds (10^26 FLOPs, etc.) as a primary regulatory hook, reinforcing that the original standards are currently still central rather than clearly obsolete.(theregister.com)

However, the prediction is about the situation by late 2026. Since that date is in the future relative to today (late 2025), we cannot yet say whether those model‑size/parameter/compute definitions will then be “largely obsolete and inapplicable.” Therefore the correct classification is inconclusive (too early) rather than right, wrong, or permanently ambiguous.