·
DataBubble
  • Home
  • Models
  • News
  • Compare
  • Boards
  • Pricing
  • About
  • Newsletter
  • Methodology
  • Contact
Latest
Bringing Value Models Back: Generative Critics for Value Modeling in LLM Reinforcement Learning8h◆Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks8h◆Distribution-Consistent Inference for Dynamic Sparse Mixture-of-Experts8h◆In RAG We Trust? Measuring Robustness of Retrieval-Augmented Generation Under Document Poisoning8h◆FrontierChallenge: Evaluating Scientific Workflow Completion8h◆IBIB: A Protocol for Measuring Enterprise AI Systems by Serving Route, Not Model Identifier8h◆AgenticGen: Reward-Guided Agentic Video Generation for Advertising8h◆Strangers to Themselves: What Language Models Say About Themselves Is Generic8h◆Omni Interaction Agent Technical Report8h◆CoGReV: A Confidence-Gated Post-Hoc Non-Monotonic Belief Revision Framework for Phishing Website Classification8h◆Bit-Flip Attacks on Vision-Language-Action Models: Action-Decoding Architecture Shapes the Vulnerability8h◆Can Foundation Models Moderate Online Content? Evaluating Instruction- vs. Example-Driven Policy Operationalization8h◆RiLM: Parameter-Efficient Language Modeling via Geodesic Decoding8h◆Tracing Computation Density in LLMs8h◆Cultural Binding Heads in Language Models8h◆Direct Diversity Optimization for Diverse Successful Trajectories in Preference Post-Training8h◆Left-Branching Transformers Excel at Right-Branching Languages: Data Shapes Word Order Preferences in Language Models8h◆Palmyra x6 Technical Report: An Agentic, Tool-Use Model Post-Trained via Anchored Supervised Fine-Tuning8h◆'Ghaib in Translation' aka Unseen Harm: Measuring Cross-Script Safety Inconsistency with 'Missed-in-Urdu' Scores in LLM Hate Speech Detection8h◆DexterSQL: Deep Schema Exploration and Rule-based Correction for Text-to-SQL Generation8h◆Bringing Value Models Back: Generative Critics for Value Modeling in LLM Reinforcement Learning8h◆Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks8h◆Distribution-Consistent Inference for Dynamic Sparse Mixture-of-Experts8h◆In RAG We Trust? Measuring Robustness of Retrieval-Augmented Generation Under Document Poisoning8h◆FrontierChallenge: Evaluating Scientific Workflow Completion8h◆IBIB: A Protocol for Measuring Enterprise AI Systems by Serving Route, Not Model Identifier8h◆AgenticGen: Reward-Guided Agentic Video Generation for Advertising8h◆Strangers to Themselves: What Language Models Say About Themselves Is Generic8h◆Omni Interaction Agent Technical Report8h◆CoGReV: A Confidence-Gated Post-Hoc Non-Monotonic Belief Revision Framework for Phishing Website Classification8h◆Bit-Flip Attacks on Vision-Language-Action Models: Action-Decoding Architecture Shapes the Vulnerability8h◆Can Foundation Models Moderate Online Content? Evaluating Instruction- vs. Example-Driven Policy Operationalization8h◆RiLM: Parameter-Efficient Language Modeling via Geodesic Decoding8h◆Tracing Computation Density in LLMs8h◆Cultural Binding Heads in Language Models8h◆Direct Diversity Optimization for Diverse Successful Trajectories in Preference Post-Training8h◆Left-Branching Transformers Excel at Right-Branching Languages: Data Shapes Word Order Preferences in Language Models8h◆Palmyra x6 Technical Report: An Agentic, Tool-Use Model Post-Trained via Anchored Supervised Fine-Tuning8h◆'Ghaib in Translation' aka Unseen Harm: Measuring Cross-Script Safety Inconsistency with 'Missed-in-Urdu' Scores in LLM Hate Speech Detection8h◆DexterSQL: Deep Schema Exploration and Rule-based Correction for Text-to-SQL Generation8h◆
News/Inference-Time Vulnerability Beyond Shallow Safety: Alignment Along Generation Trajectories
arxiv
PublishedJune 5, 2026 at 4:00 AM
—neutral

Inference-Time Vulnerability Beyond Shallow Safety: Alignment Along Generation Trajectories

Source
arxiv.orgfull article ↗
Read on arxiv→
Publisher summary· verbatim

arXiv:2606.04778v1 Announce Type: new Abstract: Safety-aligned Large Language Models (LLMs) remain vulnerable to interventions during inference that redirect generation toward harmful outputs. Recent work attributes this to shallow safety, where alignment concentrates in the first few output tokens.

Stay posted· Newsletter

A 5-min weekly brief — top movers, price watch, story of the week.

// no spam · unsubscribe one-click · free forever

Discussion
Source
↗
arxiv
Read original ↗All from arxiv →
Tags
04
#safety#large-language-models#vulnerability#alignment

No replies yet. Be first.

Source
↗
arxiv
Read original ↗All from arxiv →
Tags
04
#safety#large-language-models#vulnerability#alignment

Related coverage

More from ARXIV
arxivBringing Value Models Back: Generative Critics for Value Modeling in LLM Reinforcement Learning8harxivSubagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks8harxivDistribution-Consistent Inference for Dynamic Sparse Mixture-of-Experts8harxivIn RAG We Trust? Measuring Robustness of Retrieval-Augmented Generation Under Document Poisoning8h
The Bubble Brief
WEEKLY

Read safety insights every Tuesday — top movers, new releases, story of the week.

// no spam · unsubscribe one-click · free forever

Originally published on arxiv ↗
HomeModelsNews