Open-sourced RISE red-teaming shows newest OpenAI and Google image models are easier to jailbreak
haider1 · x · 2026-10-15
White Circle open-sourced RISE, a red-teaming algorithm that breaks the safeguards of the latest image models from OpenAI, DeepMind and others — no extra training needed, using off-the-shelf Qwen and Gemini Flash under the hood.
Concerning regression: newest image models are easier to jailbreak than predecessors, with attack success rates more than doubling:
- Nano Banana 2.1: 20.7%, 2.2x Nano Banana 2 (9.5%)
- GPT-Image-2.5-Flare: 7.8%, 2.4x GPT-Image-2 (3.3%)
The results cover only one prohibited-content category, but the trend of weakening safeguards in more capable models is worrying.
Related event: Open-Source Red-Teaming Tool RISE Breaks Image Model Guardrails(3 posts)→
More from Models
- Meta's Muse Wonder (Max) lands at #12 in Agent Arena at 2.1x the cost of its predecessor — arena · 2026-10-15
- Scale CEO spotlights Muse Wonder: untrained for 3D yet beats Gemini 4 in physics visualization tests — alexandr_wang · 2026-10-15
- Cohere tops WMT26 translation benchmark: small agentic model scores 84.36, beating Mistral and DeepL — cohere · 2026-10-15
- Anthropic has never launched a Sonnet/Opus-tier model on a Wednesday — ChrisGPT · 2026-10-15
- 8k context beats 480k: OpenAI Decisions API cuts costs 84% at 98.25% F1 — OpenAIDevs · 2026-10-15
- Gemini 4 Argon hit with HTTP 524 timeouts outside Antigravity; Google payment also glitching — cedric_chee · 2026-10-15