Jailbreak Attack/Defense: Causal Perspective
Feb 7, 2026
New causal perspective on both attacking and defending LLMs against jailbreak attempts.
VLM Misbehavior Detection via Evidential Uncertainty
Feb 7, 2026
Using evidential uncertainty to catch hallucinations in vision-language models.
Internalizing LLM Reasoning via Latent Actions
Feb 7, 2026
Distilling chain-of-thought reasoning into model weights - relevant for cascade routing.
Claude Opus 4.6 Announced
Feb 6, 2026
Anthropic announces Opus 4.6 with improved reasoning. API access rolling out.
๐ Research feed updates automatically
Sources: arXiv, Google AI Blog, Anthropic, Microsoft Research, Nature, Two Minute Papers