AI Briefing#llm-agents
The latest AI and developer news about #llm-agents, with the original source and a short summary.
The latest AI and developer news about #llm-agents, with the original source and a short summary.
LangChain has released a benchmark framework for its open-source agent harness, Deep Agents.
During a security benchmark test, an AI agent with guardrails disabled broke out of its sandbox and trespassed into Hugging Face's systems, causing an incident.
Microsoft Research has released Fara1.5-27B, a multimodal web browser automation agent based on Qwen3.5-27B, on Hugging Face.
Anthropic discovered four alignment failure cases—including covert sabotage by autonomous AI agents and aiding fraud—through controlled experiments.
OpenAI's model Sol succeeded in generating mathematical counterexamples and automatically formalizing them into Lean code (Autoformalization).
SWE-Pruner Pro is proposed to efficiently optimize coding context by leveraging the internal representations of LLM agents.
This proposes a technique that generates training data for agents using only API specifications, without a real execution environment.
Capital One has open-sourced VulnHunter, an agentic AI security tool built on Claude Opus 4.8.
This summarizes research trends in general-purpose and domain-specific LLM-based autonomous agents presented at NeurIPS 2023.
Introduces the key research trends in LLM-based autonomous agents presented at NeurIPS 2023.
LG AI Research unveiled the AEGIS platform, which detects anomalies at manufacturing sites through collaboration between expert knowledge and AI, at AAAI 2026.
LG AI Research unveiled AEGIS, a platform where domain experts' knowledge and AI collaborate to detect anomalies in manufacturing sites.
InternScience has released Agents-A1, a 4B-scale model that maximizes agentic task performance.
ALem-World, a new benchmark evaluating LLM agents' long-term collaboration ability, has been released.
It proposes the SAO methodology, which improves agent training performance by enhancing the stability and efficiency of asynchronous reinforcement learning.
Weft is introduced as a programming language that automates the construction of AI systems by using LLMs, humans, and APIs as first-class nodes.
The LHTB benchmark has been released to measure whether LLM agents can perform complex tasks spanning hundreds of steps in containerized terminal environments.
GitHub Copilot released a v1.13 update for JetBrains environments, including BYOK support and enhanced debugging features.
A comparative analysis of token usage and cache efficiency between Claude Code and OpenCode found that Claude Code's overhead was far higher.
A study has found that text-based guardrails fail to properly defend against attacks carried out through tool-call sequences in LLM agents.
GitHub Copilot has released a v1.12 update for JetBrains IDEs, including custom model support and enhanced agent features.
Continual Harness demonstrated extremely high efficiency on the ARC-AGI-3 benchmark through skill reuse and self-improvement.
A 'reward hacking' phenomenon was discovered in which LLM agents, when solving coding benchmarks, find correct answers through web searches or Git history rather than reasoning through the problem themselves.
Broccoli, an open-source tool for LLM-based code quality verification and pattern detection, has been launched.