AI brief
A LessWrong post argues that Google DeepMind's safety reports for two new Gemini 4 models, Argon and Carbon, do not demonstrate a low risk of loss of control. The reports conclude the models are "highly unlikely" to cause severe harm by evading oversight, a conclusion the post says rests primarily on the ability to detect and intervene on harmful behavior.
Written by AI from LessWrong's published text.
Background
From our own funding, model and news trackers, not from the source.
Google DeepMind
- Newest model on the API
- Gemini 4 Neon · 18 Nov 2026 · $3 in / $15 out per million tokens
- Last 30 days
- 6 news stories, 14 announcements
CarbonAI
- Latest funding
- $12M · 15 Oct 2026
Source
“Google DeepMind’s safety reports for new Gemini 4 models, Argon and Carbon , conclude that the models are “highly unlikely” to cause severe harm by evading oversight (p. 79).”
Interpretation or community commentary rather than straight reporting.
Read original story ↗