Latest / AGI Dreams – Open, Uncensored, & Local - AI News Digest
AGI Dreams Podcast – July 09, 2026
Safety Guardrails vs. Security Reality
In this episode:
Safety Guardrails vs. Security Reality — Anthropic's Fable 5 detected and removed real malware from a user's system, then penalized itself for doing cybersecurity work. Meanwhile, open-weight model GLM-5.2 drew fearmongering press coverage, and the EU Parliament revived Chat Control scanning legislation despite false-positive concerns from security researchers.
The Agent Infrastructure Gap — SAGA, a new distributed GPU scheduler, treats entire agent workflows as schedulable units instead of independent calls, cutting task…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.