Weekly / 2026

Week 36, 2026KI-Briefing KW 36/2026

OpenAI stuft GPT-6 Astra als erstes breit verfügbares Modell „Critical“ im Cyber-Risiko ein — 100 % auf ExploitBench, Zero-Days in Browser und Kernel gefunden

Issue date
Author
Stefan Braum
Period
–
Sources
10 · 9 domains
Issue
German original
German original only

Historical issues were published in German. agi.jetzt does not machine-translate them silently — the German original is linked instead.

OpenAI klassifizierte GPT-6 Astra am 3. September als „Critical“ im Cyber-Risiko — 100 % ExploitBench, Zero-Days in Browser und Kernel. Dazu: Claude Fable/Mythos 5.1, Nvidias $3,5-Mrd. bei MediaTek, EU-DSA-Pflicht für ChatGPT, Pentagon-Rollout.

Read the German original →

Specialized models & emerging approaches

From the model registry: dated entries of image, video, speech and decision models in this issue’s week. Not part of the issue’s text, and none of them is a frontier model.