6/25/2026
Memory Contagion: Cross-Temporal Propagation of Evaluator Bias via Agent Memory
Study identifying how evaluator bias propagates through agent memory systems over time, affecting LLM agents with memory-based long-term coherence.
6/25/2026
Exposing the Illusion of Erasure in Knowledge Editing for LLMs
Research on knowledge editing in LLMs revealing that edited knowledge often persists and resurfaces across model architectures, examining reliability and mechanisms.
6/25/2026
FedUP: One-Shot Federated Unlearning via Centroid-Guided Plug-in Filters
FedUP one-shot federated unlearning framework using centroid-guided plug-in filters to remove target data while preserving knowledge.
6/25/2026
FlowPipe: LLM-Enhanced Conditional Generative Flow Networks for Data Preparation Pipeline Construction
FlowPipe uses LLM-enhanced conditional generative flow networks for automated data preparation pipeline construction with improved optimization.
6/25/2026
PVF:Understanding AI Vulnerability Against SDCs
Studies AI system reliability against silent data corruptions in hardware, analyzing vulnerability of model parameters during inference.
6/25/2026
Homogeneity Bias in Open-Weight LLMs Is Robust to Decoding Hyperparameters
Maps homogeneity bias across temperature and top-p parameters in seven open-weight instruction-tuned LLMs (7-20B), finding bias robust to decoding settings.
6/25/2026
SyncLoop: A Multimodal Dual-Loop Framework for Self-Improving Mathematical Reasoning
SyncLoop dual-loop framework for self-improving multimodal LLMs with iterative refinement for mathematical reasoning without manual dataset curation.
6/25/2026
The 4/$\delta$ Bound: Designing Predictable LLM-Verifier Systems for Formal Method Guarantee
Develops formal convergence theorem for LLM-verifier systems integrating formal verification tools with LLMs for reliable software verification.
6/25/2026
A Hybrid TGN-SEAL Model for Dynamic Graph Link Prediction
Hybrid temporal graph network and SEAL model for link prediction in dynamic sparse networks capturing temporal dependencies.
6/25/2026
Multifidelity-Augmented Gaussian Process Inputs for Surrogate Modeling from Scarce Data
Multifidelity Gaussian process approach for surrogate modeling from scarce data using low and high-fidelity training samples.
6/25/2026
Breaking Data Symmetry is Needed For Generalization in Feature Learning Kernels
Theoretical analysis of grokking phenomenon in feature learning kernels, studying when data symmetry breaking enables generalization on algebraic tasks.
6/25/2026
SplitZip: Ultra Fast Lossless KV Compression for Disaggregated LLM Serving
Lossless KV cache compression technique for disaggregated LLM serving that reduces transfer bottleneck between prefill and decode workers.
6/25/2026
Defense effectiveness across architectural layers: a mechanistic evaluation of persistent memory attacks on stateful LLM agents
Systematic evaluation of six defenses against persistent memory attacks on stateful LLM agents across architectural layers with nine open-source models.
6/25/2026
Shared Doubt: Zero-Shot Cross-Lingual Confidence Estimation for Language Models
Zero-shot cross-lingual confidence estimation method for multilingual LLMs, demonstrating language-transferable confidence features without retraining.
6/25/2026
Skill-MAS: Evolving Meta-Skill for Automatic Multi-Agent Systems
Skill-MAS framework for automatic multi-agent system generation from LLMs that evolves meta-skills to balance model capability with experience retention.
6/25/2026
Approximate Structured Diffusion for Sequence Labelling
Diffusion-based approach for sequence labeling in NLP that extends linear-chain CRFs with improved expressivity through approximate structured inference.
6/25/2026
ClayBuddy: A Framework, Evaluation, & Mitigation of Coding Agent Failures
Framework and evaluation of failure modes in AI coding agents, addressing underspecification, capability errors, and mitigation strategies.
6/25/2026
EnerInfer: Energy-Aware On-Device LLM Inference
Energy-aware optimization framework for on-device LLM inference that reduces power consumption by adjusting processor frequencies while maintaining performance.
6/25/2026
Sol Video Inference Engine: Agent-Native Full-Stack Acceleration Framework for Efficient Video Generation
Framework for accelerating video diffusion model inference through instance-specific optimization strategies across different hardware configurations.
6/25/2026
Security tools inside coding agents get ignored unless we do things
Analysis of security tool adoption in AI coding agents. Examines how LLM-based agents handle security features in SDLC.
6/25/2026
Tldr we built the fastest and most compact embedded vector database in the world
LodeDB: embedded vector database optimized for local RAG. GPU-accelerated, sub-millisecond latency, compatible with LangChain and LlamaIndex.
6/25/2026
Show HN: Drive your already-logged-in Chrome from any AI agent
chrome-use: Open-source tool enabling AI agents to control real Chrome browsers with existing logins and anti-detection. Part of *-use family.
6/25/2026
Scbkr – an owner-signed responsibility-chain workbench for local LLMs
SCBKR: local LLM control framework with owner-signed responsibility chains. Governance layer for local AI ensuring user verification and audit trails.
6/25/2026
TronBrowser is an open-source, privacy-first, AI-native web browser
TronBrowser: Open-source privacy-first AI-native web browser on Ungoogled Chromium with built-in AI sidebar and agent CLI. MIT licensed.
6/25/2026
How agents are transforming work
Economic research on AI agents for knowledge work. Agents perform long-horizon delegated tasks versus short chatbot interactions.
6/25/2026
The Carwash Problem: Why Your IT Organization Isn't Ready for AI-Generated Code
Technical case study: AI desktop agent failure due to driver incompatibility creating circular dependency. Real-world deployment issue analysis.
6/25/2026
Show HN: Promptctl – Git for your AI prompts
Promptctl: Git-like version control tool for LLM prompts. Tracks, diffs, and rolls back prompts with CLI. Open-source in Go.
6/25/2026
What I'm Finding About LLM Code Style and Token Costs
Experience report on token costs and code style when using Claude for feature development. Discusses optimization patterns.
6/25/2026
Use Claude Chrome Extension from Cursor
Helper script enabling use of Claude's Chrome extension from Cursor IDE for browser tab analysis and debugging.
6/25/2026
Show HN: A durable filesystem layer for AI agents
S3-based durable filesystem layer for AI agents. Syncs memory files across platforms. Rust implementation with Python/TypeScript SDKs and CLI.
6/24/2026
The engineering bottleneck has moved
Analysis of AI prototyping tools and multiplayer collaboration challenges. Examines export compatibility and agent coordination issues in editors.
6/24/2026
Building Agent Telemetry for LLMs
Technical guide on monitoring and tracing LLM agents. Covers telemetry for multi-turn, multi-tool agent execution and error detection.
6/24/2026
Software engineers are facing an 'identity crisis bordering on depression'
Analysis of developer morale and identity issues amid AI coding tool adoption. Reports divide between 'lazy' and 'thoughtful' engineers.
6/24/2026
Anthropic says Alibaba illicitly extracted Claude AI model capabilities
Anthropic claims Alibaba extracted Claude model capabilities without authorization.
6/24/2026
Show HN: DesktopMCP – Let AI See and Operate your Linux desktop
MCP server enabling AI models to see, understand, and operate Linux desktops via XDG portals, AT-SPI, and D-Bus with 144 exposed tools.
6/24/2026
Omnigent: Open-source meta harness for agents
Open-source AI agent framework providing orchestration layer over Claude, Codex, Cursor with policy enforcement and real-time collaboration.
6/24/2026
Maturana: Hardware-isolated, zero-trust agent harness
Hardware-isolated secure agent platform using Firecracker/Hyper-V with encrypted secrets and knowledge graph.
6/24/2026
LLMs use "safety" specific neuron layers to identify vulnerabilities in code
Research analyzing how LLMs use dedicated neuron layers for safety mechanisms to identify code vulnerabilities.
6/24/2026
Same flaw, opposite verdict: what counts as a vulnerability in AI agents?
Analysis of vulnerability classification inconsistencies in AI agents—same flaw receives different verdicts across systems.
6/24/2026
OmniHai: A Java framework for all major AI providers
Java framework providing unified API for multiple AI providers (OpenAI, Claude, Gemini, etc.). Open source developer tool.
6/24/2026
Show HN: Moomacha – Zulip Native Agents Orchestrator
Open-source Zulip-native orchestrator for AI agents using chat as configuration control plane.
6/24/2026
Show HN: Drawbar – boring linear based workflows for Claude Code
Linear workflow plugin for Claude Code that accumulates knowledge base and enables sequential agent execution.
6/24/2026
The "Low Trust-Open Source" Paradox of AI Adoption in China
Research on AI infrastructure adoption barriers in China, focusing on data security trust issues in cloud and enterprise sectors.
6/24/2026
Connect Your AI Agent to Google Sheets
Tool enabling AI agents to read/write Google Sheets data. Practical LLM application for agent integration.
6/24/2026
AI enthusiasts in a race against time, AI skeptics in a race against entropy
Opinion piece comparing AI adoption mindset to infrastructure shift patterns, previews forthcoming AI analysis topics.
6/24/2026
Show HN: An LLM agent that emits typed intent
LLM agent framework emitting typed intent outputs for structured reasoning and control.
6/24/2026
Straw: Compress big infra into one md file – 99.5% LLM token reduction
Tool compressing infrastructure documentation to markdown format achieving 99.5% LLM token reduction for context efficiency.
6/24/2026
Show HN: Dspyer – self-correcting, optimizable LLM steps for DSPy and LangGraph
Tool providing typed outputs, automatic self-correction, and prompt tuning for LLM steps in DSPy and LangGraph without rewriting boilerplate.
6/24/2026
Code review powered by an LLM council
Code review system using ensemble of LLMs to evaluate pull requests. LLM application for developer workflows.
6/24/2026
Submodular Context Selection as a Pluggable Engine for LLM Agents
Research on submodular context selection as optimization engine for LLM agents, improving efficiency in information retrieval.