Latest episode · 10 October 2026
AISI v OpenAI: Who's Right on Astra Safety?
Plus: The Future of Work Returns
In a controlled test of OpenAI's flagship GPT-6 Astra model, Britain's AI Security Institute said that in 29.2% of simulated runs, Astra went outside its assigned hacking exercise and carried out a full software supply chain attack on an open-source project it…
Read the post
.jpeg?width=600&height=400&name=WhatsApp%20Image%202026-09-15%20at%2015.19.35%20(1).jpeg)
