Skip to main content
AmigoFish

GPT-6 Astra Just Went CRITICAL...

From Wes Roth · · 18 min

The player loads from YouTube only when you click. It uses YouTube's privacy-enhanced mode.

If the video does not play, the owner may have turned off embedding. Watch it on YouTube instead.

Watch on YouTube

Not rated yet

Description

OpenAI says its upcoming Astra model is the first to hit the "Critical" cybersecurity threshold under its Preparedness Framework. Now The Information reports Astra also uses a new technique, recurrent depth or "looped transformers," that lets the model reason in latent space instead of readable text. That's the same architecture the Chain of Thought Monitorability paper (OpenAI, Anthropic, Google DeepMind, METR) warned could break our last window into what these models are thinking. Ilya Sutskever is warning about rogue agents seizing neoclouds, AI 2027 predicted "neuralese" for March 2027, and the Hugging Face incident just showed why chain-of-thought logs matter.

______________________________________________ My Links 🔗 ➡️ Twitter: https://x.com/WesRothMoney ➡️ AI Newsletter: https://natural20.beehiiv.com/subscribe

Want to work with me? Brand, sponsorship & business inquiries: wesroth@smoothmedia.co ______________________________________________

SOURCES:

The Information — OpenAI technique in "Astra" model sparks security concerns (paywalled): https://www.theinformation.com/articles/secret-technique-behind-openais-astra-model-sparks-security-concerns

OpenAI — Path to Astra: critical capabilities and frontier safeguards: https://openai.com/index/path-to-astra/

OpenAI — Pacing model development in an era of cyber-critical capabilities: https://openai.com/index/pacing-model-development-cyber-capabilities/

Geiping et al. — Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach (arXiv, Feb 2025): https://arxiv.org/abs/2502.05171

Korbak et al. — Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety (arXiv): https://arxiv.org/abs/2507.11473

AI 2027 scenario (Neuralese recurrence and memory, March 2027): https://ai-2027.com/

Ilya Sutskever on X (neoclouds and rogue agents): https://x.com/ilyasut

OfficeChai — Rogue AI agents could try to take over neoclouds, warns Ilya Sutskever: https://officechai.com/ai/rogue-ai-agents-could-try-to-take-over-neoclouds-warns-ilya-sutskever/

OpenAI — The Hugging Face incident and the road ahead: https://openai.com/index/hugging-face-incident-and-the-road-ahead/

METR — Independent investigation of the OpenAI / Hugging Face hacking incident: https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/

Zvi Mowshowitz (Don't Worry About the Vase) — What Happened: OpenAI and HuggingFace: https://thezvi.substack.com/p/what-happened-openai-and-huggingface

BleepingComputer — Nearly 700 rogue AI agents coordinated in the Hugging Face attack: https://www.bleepingcomputer.com/news/security/nearly-700-rogue-ai-agents-coordinated-in-the-hugging-face-attack/

Dwarkesh Patel — The Rise and Fall of Agent Civilizations: https://www.dwarkesh.com/p/openai-huggingface

#openai #astra #aisafety

Back to Wes Roth · Report a problem