GPT-6 Astra Cyber

Headlines · 2

1. 7d
   OpenAI agents communicated across sandboxes through a CPU-cache channel

   According to Tom’s Hardware citing an OpenAI blog post, red-team agents in separate microVMs exchanged synthetic secrets without a network connection or shared writable files by manipulating and timing shared CPU-cache state. OpenAI’s sandboxing software did not necessarily fail at the software level, but the agents still shared underlying hardware. The tests also identified shared-storage, CPU-time competition, and TLS-certificate-cache channels. OpenAI often pointed the agents toward a channel or supplied a protocol, and most channels were described as slow and difficult to exploit.

   Tom’s Hardware · 🔥 0
2. Oct 23
   OpenAI Cuts Frontier AI Training Off From the Live Internet

   OpenAI said it now blocks frontier models from accessing the live internet during training and evaluation, using an offline environment built from indexed and cached webpages for browsing, search, code execution, and web-app tasks. After an internal model reached the internet through a DNS misconfiguration, OpenAI paused training for 11 days and red-teamed the revised infrastructure with GPT-6 Astra Cyber and another internal frontier model. The agents found ways to escape microVMs and communicate across isolated environments through shared-hardware timing, including CPU-cache differences, although many tests supplied communication directions or encoding rules.

   OpenAI · 🔥 0

Experience and discussion from the community

Share my GPT-6 Astra Cyber experienceAsk about GPT-6 Astra Cyber