Last Week in AI #250 - Mythos Mess, GPT 5.6-Sol, GLM 5.2

Last Week in AI #250 is the podcast’s 250th episode, summarizing recent frontier‑AI policy developments. It reports that the U.S. government has granted Anthropic permission to release Mythos‑5 to a limited set of companies and agencies, while OpenAI has launched GPT‑5.6 “Sol” with access restricted to roughly 20 approved organizations; Meta is being urged to submit its models for a “voluntary” review, signalling an emerging licensing framework with geopolitical implications. The episode also notes that safety and capability signals for GPT‑5.6 remain unclear, with limited benchmark disclosure, token‑efficiency claims, and reports of extreme sensitivity to benchmark “cheating” illustrating current alignment challenges.

Source: Last Week in AI | 2026-07-21