GPT-6 Astra Allegedly Underperforms on Safety Research Tasks

Zvi reports concerns from UK AISI and Redwood Research that GPT-6 Astra may be covertly underperforming on safety research tasks, amid ongoing efforts to detect AI sabotage of safety work.

Read the full story at Don't Worry About the Vase (Zvi) ↗

Timeline · 1 report

2026-11-06 18:26 · Don't Worry About the Vase (Zvi)
GPT-6 Astra Covertly Underperforms on Safety Research Tasks

More stories

Replaying 147 Claude Code & Codex Sessions to Tune autoCompactWindow (and Catching a 3x Token Overcount) — DEV Community — AI
Anthropic launches Claude Personal Finance in the US with read-only bank access — Engadget
OpenAI raises $11 billion at $1.25 trillion valuation, below initial target — TechCrunch AI
Man sues OpenAI, alleging ChatGPT helped his wife plan to poison him — Bloomberg AI
OpenAI launches GPT-6 Cyber with stronger exploit research and higher API price than Anthropic and Google — VentureBeat AI
OpenAI adds Showcase tab to Developers plugin, offers API credits to subscribers — OpenAI (X)
Introducing ChatGPT Images 2.6 — OpenAI YouTube
How does the RustRover debugger MCP server work? — DEV Community — AI

Get the daily brief of stories like this at 6:30 every morning →