AI Risk Wire

Models

For each AI model version: the capability levels its provider states it reached, the outside testers who tested it, and the incidents that name it.

Also: compare models side by side · statements that a capability level was crossed · providers · practices providers document · safety news.

22 of 78 model versions

ModelLevels the provider states it reachedOutside testersIncidents naming it
Gemini 4 Argon
Google DeepMind
Stealth and situational awareness TCL (Tracked Capability Level) None named None
Gemini 3.8 Flash
Google DeepMind
None stated Epoch AI None
Gemini 3.5 Audio
Google DeepMind
None stated None named None
Gemini 3.7 Flash
Google DeepMind
None stated None named None
Gemini 3.5 Flash-Lite
Google DeepMind
None stated None named None
Gemini 3.6 Flash
Google DeepMind
None stated None named None
Gemini 3.1 Flash-Lite Image
Google DeepMind · June 2026
None stated None named None
DiffusionGemma 26B A4B
Google DeepMind
None stated None named None
Gemini 3.5 Transcribe
Google DeepMind
None stated None named None
Gemini 3.5 Transcribe Live
Google DeepMind
None stated None named None
Gemini 3.5 Flash
Google DeepMind
None stated Apollo Research, Transluce, Academic None
Gemini Robotics-ER 1.6
Google DeepMind
None stated None named None
Gemini 3.1 Flash Audio
Google DeepMind
None stated None named None
Gemini 3.1 Flash-Lite
Google DeepMind
None stated Apollo Research, Redwood Research None
Gemini 3.1 Pro
Google DeepMind
None stated Apollo Research, FAR.AI, RAND (CAST / Project Canary), SecureBio and 1 more, Academic 1 incident
Supply chain attack
Veo 3
Google DeepMind
None stated None named None
Gemini 3 Flash
Google DeepMind
None stated Apollo Research, Transluce, Academic None
Gemini 3 Pro
Google DeepMind · November 2025
None stated Epoch AI, FAR.AI, RAND (CAST / Project Canary) None
Gemini 2.5 Deep Think
Google DeepMind
None stated None named None
Gemini 2.5 Pro
Google DeepMind
None stated Apollo Research, FAR.AI, RAND (CAST / Project Canary), SaferAI None
Gemini 2.0 Flash
Google DeepMind · 15 April 2025
None stated None named None
Gemini 2.0 Flash-Lite
Google DeepMind · 10 April 2025
None stated None named None