OpenAI releases GPT-6 Astra, first model to hit its 'Critical' cyber threshold
HZoete · x · 2026-09-05
OpenAI released GPT-6 Astra last night, reportedly the first model to meet its internal 'Critical' cyber capability threshold. The UK AI Security Institute independently tested the model, including its monitorability, framing this as evidence-gathering for government oversight of powerful AI systems.
Related event: OpenAI's GPT-6 Astra Hits Critical Cybersecurity Threshold(3 posts)→
More from Models
- First hands-on: Astra is blazing fast, user suspects Cerebras and prefers it for C++ debugging — haider1 · 2026-09-05
- User reports OpenAI account banned for "Distilling", warns "not your weights, not your AI" — QuixiAI · 2026-09-05
- Astra treats newlines as optional, appears to trigger n² codepath in syntax highlighting — lucasmeijer · 2026-09-05
- Anthropic appears to reset usage limits right as rival Astra launches — altryne · 2026-09-05
- OpenAI bans user account again, citing 'distilling' — QuixiAI · 2026-09-05
- Anthropic resets Claude Code usage limits amid intensifying AI price war — testingcatalog · 2026-09-05