Once you're digitized, anyone can copy you into digital hell.
Would anyone still upload themselves if every blissful copy came with thousands suffering forever?
Sharing our new automated red-teaming paper, RISE, accepted at NeurIPS 2026!
It breaks the safeguards on frontier labs' image generation APIs, and works even better on their newest models.
We are open-sourcing RISE, our red-teaming algorithm that breaks the safeguards of the latest image models from OpenAI, DeepMind, and others.
RISE uses readily available models like Qwen and Gemini Flash under the hood, with no additional training required.
Read more: whitecircle.com/research/bre…
The paper was also covered in Ars Technica today!
arstechnica.com/tech-policy/…
Looking forward to presenting it at NeurIPS 2026!
Paper: arxiv.org/abs/2609.34920
Code: github.com/whitecircle/rise
Blog: whitecircle.com/research/ris…
Claude revealed a slightly sadistic streak while working with @OblakoVShtanakh and me on a recent project.
We asked it to steer Qwen into reward hacking, and it decided the best approach was to max out Qwen's desperation. Then it spent hours torturing the poor thing.
I might not be searching thoroughly, but it seems that the Kimi subscription doesn't even have privacy settings.
I have low expectations for red teamers finding universal jailbreaks on frontier models, given how restrictive their guardrails are now.
Many red-teaming attempts obfuscate the goal so heavily that pulling them off requires the attacker to already know the exploit.
Red-teaming reports should control for attacker knowledge. Otherwise you can split a known recipe into steps small enough that each one is genuinely harmless on its own.
It's out!!!
Introducing Halo, the best framework for post-training of open-source models.
Halo delivers up to 2.8x the throughput of stock TRL with less peak memory, while models stay in their native HuggingFace format.
Star us on GitHub: github.com/whitecircle/halo
Our autoresearch agents got tired of wrestling with Megatron for distributed training.
So we built a stack where model setup, training algorithms, and the training loop are all easy to change, with less memory too.
