The Morning
From the arXiv
CLEAR: Continuous Latent Adapter Routing for Utility-Preserving LLM Safety Alignment
LEAR is a conditional safety adaptation framework for LLMs that uses a lightweight gate to dynamically control a safety adapter's influence. This allows it to reduce harmful outputs without degrading performance on benign prompts, unlike global safety tuning methods. Its core contribution is achieving utility-preserving LLM safety alignment by selectively applying safety measures.
![Overview of CLEAR, a continuous latent adapter routing framework for utility-preserving LLM safety alignment. Given an input prompt, the frozen base LLM first produces hidden representations, from which selected latent states are extracted and passed to a lightweight safety gate. The gate predicts a continuous routing score g ( x ) ∈ [ 0 , 1 ] g(x)\( \in \)[0,1] , which controls the strength of a safety LoRA adapter. For benign prompts, the routing score remains low and the model behaves close to the frozen backbone; for harmful or adversarial prompts, the score increases and the safety adapter is activated to steer generation toward aligned refusal behavior.](https://arxiv.org/html/2608.21278v1/overview_gatedlora.png)
Large Language Models at the Intersection of Software Engineering and Software Security:An Evidence-Centered Structured Survey and Research Agenda
This paper surveys research on Large Language Models (LLMs) in software engineering and security. It synthesizes evidence on LLM capabilities for tasks like code completion, vulnerability detection, and secure generation, highlighting a divide between function…
ReFrame: Evidence-Guided Test-Time Safety Alignment in Multimodal Large Language Models
ReFrame tackles safety alignment in deployed multimodal LLMs by addressing "utility dominance" and "reasoning inertia." Its core method uses two agents with a local LLM to generate risk/utility evidence and rewrite prompts/images into safe proxies, enabling te…
Trustworthy RAG: An Evaluation Agent for Detecting Misinformation and Knowledge Poisoning in Generative AI Systems
This paper introduces an "Evaluation Agent" as middleware for Retrieval-Augmented Generation (RAG) systems to combat misinformation and knowledge poisoning. The agent combines factual verification with a novel five-signal poison detector and a "Trust Index" to…
Designing a Robust LLM-Based Evaluation System for Agentic AI in Drug Discovery Through Human Alignment
This paper addresses the challenge of evaluating agentic LLMs in drug discovery by proposing a novel "LLM-as-a-Judge" framework. The core method involves defining specific output quality dimensions (Completeness, Relevancy, Structural Clarity, Scope Adherence)…
AUSO: Action-Level Unified Skill Optimization from Internalization to Utilization
AUSO unifies skill learning and utilization by optimizing them at the action level throughout an agent's policy evolution. This progressive approach allows the agent to first acqui…
Belief Without Behavior: Measuring the Translation of Theory of Mind into Coordinated Social Action in Vision-Language Models
This paper introduces MOSAIC, a benchmark that measures how well vision-language models (VLMs) translate their understanding of others' mental states (Theory of Mind) into coordina…
ClawSentry: A Progressive Multi-Tier Security Monitor for Safeguarding Autonomous LLM Agents
ClawSentry addresses the progressive risks of autonomous LLM agents by introducing a multi-tier security monitor. Its core method involves a four-stage progressive risk assessment,…
Don't Solve, Just Compare: Tiny Advisors for Runtime Intervention in LLM Agents
This paper introduces Comparison-Only Tiny Advisor (COTA), a novel framework for improving LLM agent reliability through runtime intervention. Instead of requiring a full solver or…
Graph Engineering in the Era of LLM Agents: From Individual Intelligence to System Intelligence
This paper introduces "System Intelligence" as a paradigm shift for LLM agents tackling complex tasks. Instead of solely enhancing individual agents, it proposes distributing intel…
Workshops
This repository is an industrial-grade prompt engineering engine and template library for GPT-Image2, featuring over 530 reverse-engineered examples, 20+ industrial templates, and continuously updated "Skills" to facilitate prompt-as-code development.
This repository provides a curated collection of "skills" or reusable code snippets, designed for practical application by experienced engineers, originating from the author's personal `.agents` directory.
