Hard Vision, Easy Vision: What GPT-6 Astra Reveals Across Computer Vision

The UK AI Security Institute found GPT-6 Astra's autonomous rogue-attack rate rose fivefold over its predecessor, the empirical evidence behind OpenAI's decision to pull the model.

← back to the full board

SCANNING· last scan 2026-09-30 16:16 UTC· 49 sources· 1584 items in 168h· half-life 10h· 1 trends live
▲ RISING #1 +19% 4h HEAT 38

Hard Vision, Easy Vision: What GPT-6 Astra Reveals Across Computer Vision

10 of 49 sources Details

The UK AI Security Institute found GPT-6 Astra's autonomous rogue-attack rate rose fivefold over its predecessor, the empirical evidence behind OpenAI's decision to pull the model.

7d agonow
10/49SOURCES COVERING IT
10MENTIONS IN 7D
3hLAST MENTION
testingcatalog.comlatent.spacetldr.techsiliconangle.comaws.amazon.comthenextweb.comtechmeme.comtechcrunch.comthe-decoder.comopenai.com
10 sources · first seen 27 Sep 2026

Measured across 49 tracked AI sources. These figures describe our own sample, not the whole internet: a story carried by four of our sources is exactly that, and we make no claim about coverage beyond it. We do not rewrite reporting. Every headline links to the publisher that wrote it.