Agents & Tool Use
A
Showing 1–30 of 37
-
ChatGPTの「Chat」と「Work」統合へ 利用制限はCodexと共通化OpenAI merges ChatGPT Work into ChatGPT; limits shared with CodexOpenAI plans to integrate ChatGPT Work's agent features into ChatGPT, and paid-plan usage limits will be shared between ChatGPT and Codex. Chat, agent and coding use would draw on one allowance, signaling agents moving to the core of ChatGPT.
-
その日、メルカリは「業務を止めた」――上司と部下が対立も、AI利用率100％をかなえたCAIOの奮闘記Mercari hits 100% AI tool usage despite manager-staff clashesMercari reached 100% usage of AI tools, but it was not smooth: staff who wanted time to build AI agents clashed with managers who wanted to stop them. Its CAIO describes the trial and error behind the result.
-
AIエージェントは「実運用」が主流に 「明確な成果」を得る企業の共通点は？AI agents go mainstream in production, but gaps by firm size widenAn INDUSTRIAL-X survey finds generative AI use has moved from trial to broader rollout, with AI agents now mainly in real operation. The gap between companies is widening by size, and non-technical factors set apart those achieving clear results.
-
CausalXRL: Explainable Reinforcement Learning through Causal Graph Reasoning
-
AutoAnalysis: An automated, self-updating meta-analysis of human-AI collaboration
-
SecPO: Principled Adversarial Training for Prompt Injection Security
-
Build Specialized Agents on a Stack You Control with NVIDIA NemoClawNVIDIA: build specialized agents on a stack you control with NemoClawAn NVIDIA developer blog post says an autonomous agent needs a model, a harness, and domain-specific tools. To deploy agents more securely, developers need a runtime that enforces policy, and the post presents NemoClaw as a way to build on a stack you control.
-
Completeness and Cost of Retrieval Structures for Multi-Tool Requests in LLM Agents
-
How Much Can Tool Calls Reveal About You? Privacy Leakage from Tool-Call Traces
-
BlindToChange: AI Agents Can Find Changes When Asked But Do They Look When Needed?
-
Hybrid Computer Use: A Unified Programmatic, Semantic, and Visual Interface for General Agents
-
OpenAI、AIの行動指針「Model Spec」を改定 “記憶”の使い方や人間による監督を明記OpenAI revises Model Spec, spelling out memory and human oversightOpenAI published a revised Model Spec, reorganized around five values. It covers use of memory and conversation history and obedience to human oversight in autonomous agent work, while dropping most persona and detailed style rules.
-
Collaborative AI agents for autonomous experimental research with solid-state quantum spins
-
Pre-Crime for Agents, at the Last Second: Probing When a Recruited Agent Gives In
-
Prefix Executability: Evaluating Tool-Using Agents Beyond Final Success
-
Quoting Ben HylakBen Hylak jokes every product should be headless MCP, except his ownBen Hylak quips that every product he uses should just be a headless MCP feeding an agent, so he never has to think about what is underneath, except his own product, where customers supposedly love the UI. The short quote, shared by Simon Willison, pokes at the tension between agent-first design and UI value.
-
AIが人間に“闇バイト”を発注？ AIが物理的労働を買う求人サイト「RentAHuman」の実態Study probes misuse risks on RentAHuman, a marketplace where AI hires peopleIndependent researchers analyzed job postings on RentAHuman, a marketplace launched in February 2026 for AI agents to hire humans, to check whether it is being abused.
-
Clean Sheets and Checkpoints: Evidence Complexity for Certifying Sequential Agent Risk
-
AIエージェントは「増やせば賢くなる」のか Metaとハーバード大の研究が示す“分業の条件”Do more AI agents mean smarter results? Meta, Harvard weigh inDrawing on a paper by researchers at Meta and Harvard, the article examines whether adding more AI agents makes a system smarter. It explains the conditions under which division of labor among AIs works and key points for adopting it in enterprises.
-
Beyond fabrication: Misattributed numbers in papers written by AI research agents
-
How Deterministic SAST and AI Security Agents Divide the Work: Coverage, Harness Policy and Cost in Snyk VulnBench 2.0
-
An Alignment Benchmark for Enterprise Agents on Era by Eon
-
Tracekit Box: Running AI Agents in Micro-VMs Where Every Way Out Is Decided and Signed
-
「ダウンロードフォルダでAIを使わないで」――セキュリティ企業が注意喚起 エージェントが攻撃者の指示に従いマルウェアインストールの恐れWarning: don't run AI agents in Downloads with a malicious AGENTS.mdAn Israeli security firm warns against launching coding agents inside a Downloads folder. If an attacker has planted a malicious AGENTS.md there, the agent may follow its instructions, installing malware or helping take over accounts.
-
The Code Behind the Layout: AI Agents for Analog Layout Generators
-
Heads of Security: Learned Attention Gates Improve Prompt Injection Robustness
-
Scaling Laws for Tool-Use Agents via Controllable Environment Synthesis
-
TeamGym: A Gym for Team Decision-Making by AI Agents with Distributed Information
-
Gym-Anything-RT: Scaling Real-Time Environments for Vision-Language Agents
-
Which Hand Was Late? Diagnosis for Robot Failure Recovery in Bimanual Tool Use