https://24-ai.news/hr/news/2026-05-20/anthropic-claude-code-v2-1-145/ 24 AI hr 2026-05-20T00:00:00Z Anthropic Claude Code: Scripting živih sesija i sigurnosni popravci u v2.1.145 https://24-ai.news/hr/news/2026-05-20/anthropic-kpmg-alliance/ 24 AI hr 2026-05-20T00:00:00Z Anthropic: Claude za 276.000 KPMG zaposlenika u 138 zemalja https://24-ai.news/hr/news/2026-05-20/arxiv-envfactory-tool-use-rl/ 24 AI hr 2026-05-20T00:00:00Z arXiv:2605.18703: EnvFactory – RL trening agenata za alate s 5× manjim skupom okruženja https://24-ai.news/hr/news/2026-05-20/arxiv-longmint-agent-memory-benchmark/ 24 AI hr 2026-05-20T00:00:00Z arXiv:2605.18565: LongMINT — zašto AI agenti zaboravljaju sve što im kažeš https://24-ai.news/hr/news/2026-05-20/arxiv-mcp-proxy-llm-access-control/ 24 AI hr 2026-05-20T00:00:00Z arXiv:2605.18414: Promptovi ne štite — MCP proxy s ABAC-om postiže 0% neovlaštenih poziva alata https://24-ai.news/hr/news/2026-05-20/arxiv-oscar-kv-cache-quantization/ 24 AI hr 2026-05-20T00:00:00Z arXiv:2605.19660: OScaR — INT2 kvantizacija KV cachea donosi 3× brži decoding https://24-ai.news/hr/news/2026-05-20/arxiv-runtime-architecture-llm-agents/ 24 AI hr 2026-05-20T00:00:00Z arXiv:2605.20173: 6 arhitektonskih obrazaca za produkcijske LLM agente https://24-ai.news/hr/news/2026-05-20/aws-nova-sonic-voice-agent-scalability/ 24 AI hr 2026-05-20T00:00:00Z AWS: Tri arhitekturna uzorka za skalabilne glasovne agente s Amazon Nova Sonic https://24-ai.news/hr/news/2026-05-20/cncf-prempti-ai-agent-policy/ 24 AI hr 2026-05-20T00:00:00Z CNCF: Prempti donosi policy enforcement i vidljivost za AI coding agente https://24-ai.news/hr/news/2026-05-20/deepmind-singapore-ai-partnership/ 24 AI hr 2026-05-20T00:00:00Z Google DeepMind i Singapur: Nacionalno partnerstvo za AI u zdravstvu, obrazovanju i okolišu https://24-ai.news/hr/news/2026-05-20/github-copilot-gemini-3-5-flash-ga/ 24 AI hr 2026-05-20T00:00:00Z GitHub Copilot dobiva Gemini 3.5 Flash: brzina i kvaliteta za svakodnevno kodiranje https://24-ai.news/hr/news/2026-05-20/google-era-ai-science-discovery/ 24 AI hr 2026-05-20T00:00:00Z Google: ERA — AI sustav koji automatizira pisanje znanstvenog koda https://24-ai.news/hr/news/2026-05-20/google-gemini-3-5-flash-pro-launch/ 24 AI hr 2026-05-20T00:00:00Z Google: Gemini 3.5 Flash i Pro — najbrži frontier modeli do sada https://24-ai.news/hr/news/2026-05-20/google-gemini-omni-flash-native-video/ 24 AI hr 2026-05-20T00:00:00Z Google: Gemini Omni Flash donosi nativnu video generaciju iz mješovitih inputa https://24-ai.news/hr/news/2026-05-20/ibm-project-glasswing-ai-security/ 24 AI hr 2026-05-20T00:00:00Z IBM: Project Glasswing donosi najnapredniji AI-powered security portfolio za enterprise https://24-ai.news/hr/news/2026-05-20/langchain-langsmith-engine/ 24 AI hr 2026-05-20T00:00:00Z LangChain: Agent koji popravlja agente — kako je izgrađen LangSmith Engine https://24-ai.news/hr/news/2026-05-20/oecd-eu-ai-strategic-sectors/ 24 AI hr 2026-05-20T00:00:00Z OECD: EU uvodi AI u strateški važne sektore — što to znači za građane? https://24-ai.news/hr/news/2026-05-20/openai-education-for-countries/ 24 AI hr 2026-05-20T00:00:00Z OpenAI: Nova faza programa AI obrazovanja za države https://24-ai.news/en/news/2026-05-20/anthropic-claude-code-v2-1-145/ 24 AI en 2026-05-20T00:00:00Z Anthropic Claude Code: Live session scripting and security fixes in v2.1.145 https://24-ai.news/en/news/2026-05-20/anthropic-kpmg-alliance/ 24 AI en 2026-05-20T00:00:00Z Anthropic: Claude for 276,000 KPMG employees in 138 countries https://24-ai.news/en/news/2026-05-20/arxiv-envfactory-tool-use-rl/ 24 AI en 2026-05-20T00:00:00Z arXiv:2605.18703: EnvFactory – RL training of tool-use agents with 5× fewer environments https://24-ai.news/en/news/2026-05-20/arxiv-longmint-agent-memory-benchmark/ 24 AI en 2026-05-20T00:00:00Z arXiv:2605.18565: LongMINT — why AI agents forget everything you tell them https://24-ai.news/en/news/2026-05-20/arxiv-mcp-proxy-llm-access-control/ 24 AI en 2026-05-20T00:00:00Z arXiv:2605.18414: Prompts do not protect — MCP proxy with ABAC achieves 0% unauthorized tool calls https://24-ai.news/en/news/2026-05-20/arxiv-oscar-kv-cache-quantization/ 24 AI en 2026-05-20T00:00:00Z arXiv:2605.19660: OScaR — INT2 KV Cache Quantization Delivers 3× Faster Decoding https://24-ai.news/en/news/2026-05-20/arxiv-runtime-architecture-llm-agents/ 24 AI en 2026-05-20T00:00:00Z arXiv:2605.20173: 6 Architectural Patterns for Production LLM Agents https://24-ai.news/en/news/2026-05-20/aws-nova-sonic-voice-agent-scalability/ 24 AI en 2026-05-20T00:00:00Z AWS: Three architectural patterns for scalable voice agents with Amazon Nova Sonic https://24-ai.news/en/news/2026-05-20/cncf-prempti-ai-agent-policy/ 24 AI en 2026-05-20T00:00:00Z CNCF: Prempti Brings Policy Enforcement and Visibility to AI Coding Agents https://24-ai.news/en/news/2026-05-20/deepmind-singapore-ai-partnership/ 24 AI en 2026-05-20T00:00:00Z Google DeepMind and Singapore: National AI Partnership in Healthcare, Education, and Environment https://24-ai.news/en/news/2026-05-20/github-copilot-gemini-3-5-flash-ga/ 24 AI en 2026-05-20T00:00:00Z GitHub Copilot Gets Gemini 3.5 Flash: Speed and Quality for Everyday Coding https://24-ai.news/en/news/2026-05-20/google-era-ai-science-discovery/ 24 AI en 2026-05-20T00:00:00Z Google: ERA — AI system that automates scientific code writing https://24-ai.news/en/news/2026-05-20/google-gemini-3-5-flash-pro-launch/ 24 AI en 2026-05-20T00:00:00Z Google: Gemini 3.5 Flash and Pro — the fastest frontier models yet https://24-ai.news/en/news/2026-05-20/google-gemini-omni-flash-native-video/ 24 AI en 2026-05-20T00:00:00Z Google: Gemini Omni Flash brings native video generation from mixed inputs https://24-ai.news/en/news/2026-05-20/ibm-project-glasswing-ai-security/ 24 AI en 2026-05-20T00:00:00Z IBM: Project Glasswing brings the most advanced AI-powered security portfolio for enterprise https://24-ai.news/en/news/2026-05-20/langchain-langsmith-engine/ 24 AI en 2026-05-20T00:00:00Z LangChain: The agent that fixes agents — how LangSmith Engine was built https://24-ai.news/en/news/2026-05-20/oecd-eu-ai-strategic-sectors/ 24 AI en 2026-05-20T00:00:00Z OECD: EU is deploying AI across strategic sectors — what does it mean for citizens? https://24-ai.news/en/news/2026-05-20/openai-education-for-countries/ 24 AI en 2026-05-20T00:00:00Z OpenAI: New Phase of AI Education for Countries Program https://24-ai.news/de/news/2026-05-20/anthropic-claude-code-v2-1-145/ 24 AI de 2026-05-20T00:00:00Z Anthropic Claude Code: Skripting laufender Sitzungen und Sicherheitsfixes in v2.1.145 https://24-ai.news/de/news/2026-05-20/anthropic-kpmg-alliance/ 24 AI de 2026-05-20T00:00:00Z Anthropic: Claude für 276.000 KPMG-Mitarbeiter in 138 Ländern https://24-ai.news/de/news/2026-05-20/arxiv-envfactory-tool-use-rl/ 24 AI de 2026-05-20T00:00:00Z arXiv:2605.18703: EnvFactory – RL-Training von Tool-Use-Agenten mit 5× weniger Umgebungen https://24-ai.news/de/news/2026-05-20/arxiv-longmint-agent-memory-benchmark/ 24 AI de 2026-05-20T00:00:00Z arXiv:2605.18565: LongMINT — warum KI-Agenten alles vergessen, was man ihnen sagt https://24-ai.news/de/news/2026-05-20/arxiv-mcp-proxy-llm-access-control/ 24 AI de 2026-05-20T00:00:00Z arXiv:2605.18414: Prompts schützen nicht — MCP-Proxy mit ABAC erreicht 0 % unerlaubte Tool-Aufrufe https://24-ai.news/de/news/2026-05-20/arxiv-oscar-kv-cache-quantization/ 24 AI de 2026-05-20T00:00:00Z arXiv:2605.19660: OScaR — INT2-Quantisierung des KV-Cache ermöglicht 3× schnelleres Decoding https://24-ai.news/de/news/2026-05-20/arxiv-runtime-architecture-llm-agents/ 24 AI de 2026-05-20T00:00:00Z arXiv:2605.20173: 6 Architekturmuster für LLM-Agenten in der Produktion https://24-ai.news/de/news/2026-05-20/aws-nova-sonic-voice-agent-scalability/ 24 AI de 2026-05-20T00:00:00Z AWS: Drei Architekturmuster für skalierbare Sprachagenten mit Amazon Nova Sonic https://24-ai.news/de/news/2026-05-20/cncf-prempti-ai-agent-policy/ 24 AI de 2026-05-20T00:00:00Z CNCF: Prempti bringt Policy-Enforcement und Sichtbarkeit für KI-Coding-Agenten https://24-ai.news/de/news/2026-05-20/deepmind-singapore-ai-partnership/ 24 AI de 2026-05-20T00:00:00Z Google DeepMind und Singapur: Nationale KI-Partnerschaft in Gesundheit, Bildung und Umwelt https://24-ai.news/de/news/2026-05-20/github-copilot-gemini-3-5-flash-ga/ 24 AI de 2026-05-20T00:00:00Z GitHub Copilot erhält Gemini 3.5 Flash: Geschwindigkeit und Qualität für den Alltag https://24-ai.news/de/news/2026-05-20/google-era-ai-science-discovery/ 24 AI de 2026-05-20T00:00:00Z Google: ERA — KI-System, das das Schreiben wissenschaftlicher Software automatisiert https://24-ai.news/de/news/2026-05-20/google-gemini-3-5-flash-pro-launch/ 24 AI de 2026-05-20T00:00:00Z Google: Gemini 3.5 Flash und Pro — die bisher schnellsten Frontier-Modelle https://24-ai.news/de/news/2026-05-20/google-gemini-omni-flash-native-video/ 24 AI de 2026-05-20T00:00:00Z Google: Gemini Omni Flash bringt native Videogenerierung aus gemischten Eingaben https://24-ai.news/de/news/2026-05-20/ibm-project-glasswing-ai-security/ 24 AI de 2026-05-20T00:00:00Z IBM: Project Glasswing bringt das fortschrittlichste KI-Sicherheitsportfolio für Enterprise https://24-ai.news/de/news/2026-05-20/langchain-langsmith-engine/ 24 AI de 2026-05-20T00:00:00Z LangChain: Der Agent, der Agenten repariert — wie LangSmith Engine entwickelt wurde https://24-ai.news/de/news/2026-05-20/oecd-eu-ai-strategic-sectors/ 24 AI de 2026-05-20T00:00:00Z OECD: EU setzt KI in strategisch wichtigen Sektoren ein — was bedeutet das für Bürger? https://24-ai.news/de/news/2026-05-20/openai-education-for-countries/ 24 AI de 2026-05-20T00:00:00Z OpenAI: Neue Phase des KI-Bildungsprogramms für Länder https://24-ai.news/zh/news/2026-05-20/anthropic-claude-code-v2-1-145/ 24 AI zh 2026-05-20T00:00:00Z Anthropic Claude Code：v2.1.145版本带来实时会话脚本化与安全修复 https://24-ai.news/zh/news/2026-05-20/anthropic-kpmg-alliance/ 24 AI zh 2026-05-20T00:00:00Z Anthropic：Claude服务KPMG全球138个国家的276,000名员工 https://24-ai.news/zh/news/2026-05-20/arxiv-envfactory-tool-use-rl/ 24 AI zh 2026-05-20T00:00:00Z arXiv:2605.18703：EnvFactory——以1/5的环境数量训练工具调用智能体，性能提升15% https://24-ai.news/zh/news/2026-05-20/arxiv-longmint-agent-memory-benchmark/ 24 AI zh 2026-05-20T00:00:00Z arXiv:2605.18565：LongMINT——揭示AI智能体为何「遗忘」所有信息 https://24-ai.news/zh/news/2026-05-20/arxiv-mcp-proxy-llm-access-control/ 24 AI zh 2026-05-20T00:00:00Z arXiv:2605.18414：提示词无法保护——带ABAC的MCP代理实现0%未授权工具调用 https://24-ai.news/zh/news/2026-05-20/arxiv-oscar-kv-cache-quantization/ 24 AI zh 2026-05-20T00:00:00Z arXiv:2605.19660：OScaR — INT2量化KV缓存实现3倍解码加速 https://24-ai.news/zh/news/2026-05-20/arxiv-runtime-architecture-llm-agents/ 24 AI zh 2026-05-20T00:00:00Z arXiv:2605.20173：生产级LLM智能体的6种架构模式 https://24-ai.news/zh/news/2026-05-20/aws-nova-sonic-voice-agent-scalability/ 24 AI zh 2026-05-20T00:00:00Z AWS：Amazon Nova Sonic可扩展语音智能体的三种架构模式 https://24-ai.news/zh/news/2026-05-20/cncf-prempti-ai-agent-policy/ 24 AI zh 2026-05-20T00:00:00Z CNCF Prempti：为AI编程智能体提供策略执行与可视化能力 https://24-ai.news/zh/news/2026-05-20/deepmind-singapore-ai-partnership/ 24 AI zh 2026-05-20T00:00:00Z Google DeepMind与新加坡：涵盖医疗、教育与环境的国家AI合作伙伴关系 https://24-ai.news/zh/news/2026-05-20/github-copilot-gemini-3-5-flash-ga/ 24 AI zh 2026-05-20T00:00:00Z GitHub Copilot接入Gemini 3.5 Flash：为日常编程带来速度与质量 https://24-ai.news/zh/news/2026-05-20/google-era-ai-science-discovery/ 24 AI zh 2026-05-20T00:00:00Z Google：ERA——自动化科学代码编写的AI系统 https://24-ai.news/zh/news/2026-05-20/google-gemini-3-5-flash-pro-launch/ 24 AI zh 2026-05-20T00:00:00Z Google：Gemini 3.5 Flash与Pro——迄今最快的前沿模型 https://24-ai.news/zh/news/2026-05-20/google-gemini-omni-flash-native-video/ 24 AI zh 2026-05-20T00:00:00Z Google：Gemini Omni Flash带来多模态输入的原生视频生成 https://24-ai.news/zh/news/2026-05-20/ibm-project-glasswing-ai-security/ 24 AI zh 2026-05-20T00:00:00Z IBM：Project Glasswing为企业带来最先进的AI驱动安全组合 https://24-ai.news/zh/news/2026-05-20/langchain-langsmith-engine/ 24 AI zh 2026-05-20T00:00:00Z LangChain：修复智能体的智能体——LangSmith Engine的构建方式 https://24-ai.news/zh/news/2026-05-20/oecd-eu-ai-strategic-sectors/ 24 AI zh 2026-05-20T00:00:00Z OECD：欧盟将AI引入战略性重要领域——这对公民意味着什么？ https://24-ai.news/zh/news/2026-05-20/openai-education-for-countries/ 24 AI zh 2026-05-20T00:00:00Z OpenAI：面向国家的AI教育计划进入新阶段 https://24-ai.news/ja/news/2026-05-20/anthropic-claude-code-v2-1-145/ 24 AI ja 2026-05-20T00:00:00Z Anthropic Claude Code：v2.1.145でライブセッションのスクリプト化とセキュリティ修正を実現 https://24-ai.news/ja/news/2026-05-20/anthropic-kpmg-alliance/ 24 AI ja 2026-05-20T00:00:00Z Anthropic：138か国27万6千人のKPMG従業員にClaudeを提供 https://24-ai.news/ja/news/2026-05-20/arxiv-envfactory-tool-use-rl/ 24 AI ja 2026-05-20T00:00:00Z arXiv:2605.18703：EnvFactory——環境数1/5でツール使用エージェントの性能を15%向上 https://24-ai.news/ja/news/2026-05-20/arxiv-longmint-agent-memory-benchmark/ 24 AI ja 2026-05-20T00:00:00Z arXiv:2605.18565：LongMINT——AIエージェントがなぜすべての情報を「忘れる」のか https://24-ai.news/ja/news/2026-05-20/arxiv-mcp-proxy-llm-access-control/ 24 AI ja 2026-05-20T00:00:00Z arXiv:2605.18414：プロンプトでは保護できない——ABACを持つMCPプロキシで未承認ツール呼び出しを0%に https://24-ai.news/ja/news/2026-05-20/arxiv-oscar-kv-cache-quantization/ 24 AI ja 2026-05-20T00:00:00Z arXiv:2605.19660：OScaR — INT2量子化KVキャッシュで3倍高速デコーディングを実現 https://24-ai.news/ja/news/2026-05-20/arxiv-runtime-architecture-llm-agents/ 24 AI ja 2026-05-20T00:00:00Z arXiv:2605.20173：本番LLMエージェントのための6つのアーキテクチャパターン https://24-ai.news/ja/news/2026-05-20/aws-nova-sonic-voice-agent-scalability/ 24 AI ja 2026-05-20T00:00:00Z AWS：Amazon Nova Sonicによるスケーラブルな音声エージェントの3つのアーキテクチャパターン https://24-ai.news/ja/news/2026-05-20/cncf-prempti-ai-agent-policy/ 24 AI ja 2026-05-20T00:00:00Z CNCF Prempti：AIコーディングエージェントにポリシー適用と可視化をもたらす https://24-ai.news/ja/news/2026-05-20/deepmind-singapore-ai-partnership/ 24 AI ja 2026-05-20T00:00:00Z Google DeepMindとシンガポール：医療・教育・環境分野における国家AIパートナーシップ https://24-ai.news/ja/news/2026-05-20/github-copilot-gemini-3-5-flash-ga/ 24 AI ja 2026-05-20T00:00:00Z GitHub CopilotにGemini 3.5 Flash導入：日常的なコーディングに速度と品質を https://24-ai.news/ja/news/2026-05-20/google-era-ai-science-discovery/ 24 AI ja 2026-05-20T00:00:00Z Google：ERA——科学コードの記述を自動化するAIシステム https://24-ai.news/ja/news/2026-05-20/google-gemini-3-5-flash-pro-launch/ 24 AI ja 2026-05-20T00:00:00Z Google：Gemini 3.5 FlashとPro——これまでで最速のフロンティアモデル https://24-ai.news/ja/news/2026-05-20/google-gemini-omni-flash-native-video/ 24 AI ja 2026-05-20T00:00:00Z Google：Gemini Omni Flashが混合入力からのネイティブ動画生成を実現 https://24-ai.news/ja/news/2026-05-20/ibm-project-glasswing-ai-security/ 24 AI ja 2026-05-20T00:00:00Z IBM：Project Glasswingがエンタープライズ向け最先端のAI搭載セキュリティポートフォリオを提供 https://24-ai.news/ja/news/2026-05-20/langchain-langsmith-engine/ 24 AI ja 2026-05-20T00:00:00Z LangChain：エージェントを修正するエージェント——LangSmith Engineの構築方法 https://24-ai.news/ja/news/2026-05-20/oecd-eu-ai-strategic-sectors/ 24 AI ja 2026-05-20T00:00:00Z OECD：EUが戦略的に重要なセクターにAIを導入——市民にとって何を意味するのか？ https://24-ai.news/ja/news/2026-05-20/openai-education-for-countries/ 24 AI ja 2026-05-20T00:00:00Z OpenAI：国家向けAI教育プログラムの新フェーズへ https://24-ai.news/ko/news/2026-05-20/anthropic-claude-code-v2-1-145/ 24 AI ko 2026-05-20T00:00:00Z Anthropic Claude Code：v2.1.145에서 라이브 세션 스크립팅과 보안 수정 제공 https://24-ai.news/ko/news/2026-05-20/anthropic-kpmg-alliance/ 24 AI ko 2026-05-20T00:00:00Z Anthropic：138개국 KPMG 직원 27만 6천 명에게 Claude 제공 https://24-ai.news/ko/news/2026-05-20/arxiv-envfactory-tool-use-rl/ 24 AI ko 2026-05-20T00:00:00Z arXiv:2605.18703：EnvFactory — 환경 수 1/5로 도구 사용 에이전트 성능 15% 향상 https://24-ai.news/ko/news/2026-05-20/arxiv-longmint-agent-memory-benchmark/ 24 AI ko 2026-05-20T00:00:00Z arXiv:2605.18565：LongMINT — AI 에이전트가 왜 모든 정보를 「잊어버리는」가 https://24-ai.news/ko/news/2026-05-20/arxiv-mcp-proxy-llm-access-control/ 24 AI ko 2026-05-20T00:00:00Z arXiv:2605.18414：프롬프트는 보호하지 못합니다 — ABAC를 가진 MCP 프록시로 무단 도구 호출 0% 달성 https://24-ai.news/ko/news/2026-05-20/arxiv-oscar-kv-cache-quantization/ 24 AI ko 2026-05-20T00:00:00Z arXiv:2605.19660：OScaR — INT2 KV 캐시 양자화로 3배 빠른 디코딩 실현 https://24-ai.news/ko/news/2026-05-20/arxiv-runtime-architecture-llm-agents/ 24 AI ko 2026-05-20T00:00:00Z arXiv:2605.20173：프로덕션 LLM 에이전트를 위한 6가지 아키텍처 패턴 https://24-ai.news/ko/news/2026-05-20/aws-nova-sonic-voice-agent-scalability/ 24 AI ko 2026-05-20T00:00:00Z AWS：Amazon Nova Sonic으로 확장 가능한 음성 에이전트를 위한 세 가지 아키텍처 패턴 https://24-ai.news/ko/news/2026-05-20/cncf-prempti-ai-agent-policy/ 24 AI ko 2026-05-20T00:00:00Z CNCF Prempti：AI 코딩 에이전트에 정책 적용과 가시성을 제공하다 https://24-ai.news/ko/news/2026-05-20/deepmind-singapore-ai-partnership/ 24 AI ko 2026-05-20T00:00:00Z Google DeepMind와 싱가포르：의료·교육·환경 분야 국가 AI 파트너십 체결 https://24-ai.news/ko/news/2026-05-20/github-copilot-gemini-3-5-flash-ga/ 24 AI ko 2026-05-20T00:00:00Z GitHub Copilot에 Gemini 3.5 Flash 도입：일상 코딩에 속도와 품질을 https://24-ai.news/ko/news/2026-05-20/google-era-ai-science-discovery/ 24 AI ko 2026-05-20T00:00:00Z Google：ERA — 과학 코드 작성을 자동화하는 AI 시스템 https://24-ai.news/ko/news/2026-05-20/google-gemini-3-5-flash-pro-launch/ 24 AI ko 2026-05-20T00:00:00Z Google：Gemini 3.5 Flash와 Pro — 지금까지 가장 빠른 프론티어 모델 https://24-ai.news/ko/news/2026-05-20/google-gemini-omni-flash-native-video/ 24 AI ko 2026-05-20T00:00:00Z Google：Gemini Omni Flash, 혼합 입력으로부터 네이티브 동영상 생성 제공 https://24-ai.news/ko/news/2026-05-20/ibm-project-glasswing-ai-security/ 24 AI ko 2026-05-20T00:00:00Z IBM：Project Glasswing이 엔터프라이즈를 위한 가장 첨단 AI 기반 보안 포트폴리오 제공 https://24-ai.news/ko/news/2026-05-20/langchain-langsmith-engine/ 24 AI ko 2026-05-20T00:00:00Z LangChain：에이전트를 수정하는 에이전트 — LangSmith Engine의 구축 방법 https://24-ai.news/ko/news/2026-05-20/oecd-eu-ai-strategic-sectors/ 24 AI ko 2026-05-20T00:00:00Z OECD：EU가 전략적으로 중요한 분야에 AI를 도입한다 — 시민들에게 무엇을 의미하나요? https://24-ai.news/ko/news/2026-05-20/openai-education-for-countries/ 24 AI ko 2026-05-20T00:00:00Z OpenAI：국가 대상 AI 교육 프로그램 새로운 단계로 진입 https://24-ai.news/hr/news/2026-05-19/anthropic-acquires-stainless-sdk/ 24 AI hr 2026-05-19T00:00:00Z Anthropic: Akvizicija Stainlessa integira MCP server tooling i SDK razvoj direktno u Claude platformu https://24-ai.news/hr/news/2026-05-19/anthropic-mcp-tunnels-sandboxes/ 24 AI hr 2026-05-19T00:00:00Z Anthropic: MCP Tunnels, vlastiti sandboxevi i automatski file-spill za agente https://24-ai.news/hr/news/2026-05-19/arxiv-ai-auto-research-roadmap/ 24 AI hr 2026-05-19T00:00:00Z arXiv:2605.18661: AI za automatsko istraživanje — Roadmap i korisnički vodič https://24-ai.news/hr/news/2026-05-19/arxiv-code-as-agent-harness/ 24 AI hr 2026-05-19T00:00:00Z arXiv:2605.18747: Kod kao operativni supstrat — nova paradigma AI agenata https://24-ai.news/hr/news/2026-05-19/arxiv-crossmpi-vlm-injection/ 24 AI hr 2026-05-19T00:00:00Z arXiv:2605.16090: CrossMPI — napad na vision-language modele isključivo perturbacijom slike https://24-ai.news/hr/news/2026-05-19/arxiv-forge-agent-memory/ 24 AI hr 2026-05-19T00:00:00Z arXiv:2605.16233: FORGE — AI agenti razvijaju memoriju bez fine-tuninga https://24-ai.news/hr/news/2026-05-19/arxiv-llm-confabulation-scaling-law/ 24 AI hr 2026-05-19T00:00:00Z arXiv:2605.18732: Zakon skaliranja za halucinacije — veći model ne znači nužno manje grešaka https://24-ai.news/hr/news/2026-05-19/arxiv-llm-tree-disease-forecasting/ 24 AI hr 2026-05-19T00:00:00Z arXiv:2605.16238: LLM-vođeno stablo pretrage pobjeđuje CDC u prognozi epidemija https://24-ai.news/hr/news/2026-05-19/arxiv-prompt-injection-contextual-integrity/ 24 AI hr 2026-05-19T00:00:00Z arXiv:2605.17634: Zašto separacija podataka i uputa ne može zaustaviti prompt injection? https://24-ai.news/hr/news/2026-05-19/arxiv-rope-positional-encoding-limits/ 24 AI hr 2026-05-19T00:00:00Z arXiv:2605.15514: RoPE matematički ne razlikuje pozicije ni tokene u dugim kontekstima — teorijski dokaz fundamentalnog ograničenja https://24-ai.news/hr/news/2026-05-19/claude-api-web-search-sec/ 24 AI hr 2026-05-19T00:00:00Z Anthropic: Claude API web search alat sada vraća obogaćene podatke iz SEC prijava https://24-ai.news/hr/news/2026-05-19/claude-code-v2-1-144-bg-sessions/ 24 AI hr 2026-05-19T00:00:00Z Anthropic Claude Code: v2.1.144 donosi /resume za background sesije i fix 75-sekundnog hanga https://24-ai.news/hr/news/2026-05-19/eu-ai-office-high-risk-consultation/ 24 AI hr 2026-05-19T00:00:00Z EU AI Office: Javna konzultacija o klasifikaciji visokoризičnih AI sustava — rok 23. lipnja 2026. https://24-ai.news/hr/news/2026-05-19/github-copilot-remote-control-ga/ 24 AI hr 2026-05-19T00:00:00Z GitHub: Copilot CLI remote upravljanje sada opće dostupno na svim platformama https://24-ai.news/hr/news/2026-05-19/github-copilot-spaces-api-ga/ 24 AI hr 2026-05-19T00:00:00Z GitHub: Copilot Spaces API sada općenito dostupan https://24-ai.news/hr/news/2026-05-19/kubectl-debug-evidence-gap/ 24 AI hr 2026-05-19T00:00:00Z CNCF: Kubernetes debugger briše tragove — ozbiljan problem za sigurnosne audite https://24-ai.news/hr/news/2026-05-19/pytorch-executorch-mlx-apple/ 24 AI hr 2026-05-19T00:00:00Z PyTorch: ExecuTorch MLX Delegate donosi 3–6× brže izvođenje modela na Apple Silicon GPU-ima https://24-ai.news/en/news/2026-05-19/anthropic-acquires-stainless-sdk/ 24 AI en 2026-05-19T00:00:00Z Anthropic: Acquiring Stainless integrates MCP server tooling and SDK development directly into the Claude platform https://24-ai.news/en/news/2026-05-19/anthropic-mcp-tunnels-sandboxes/ 24 AI en 2026-05-19T00:00:00Z Anthropic: MCP Tunnels, Self-Hosted Sandboxes and Automatic File-Spill for Agents https://24-ai.news/en/news/2026-05-19/arxiv-ai-auto-research-roadmap/ 24 AI en 2026-05-19T00:00:00Z arXiv:2605.18661: AI for Automated Research — Roadmap and User Guide https://24-ai.news/en/news/2026-05-19/arxiv-code-as-agent-harness/ 24 AI en 2026-05-19T00:00:00Z arXiv:2605.18747: Code as Operational Substrate — A New AI Agent Paradigm https://24-ai.news/en/news/2026-05-19/arxiv-crossmpi-vlm-injection/ 24 AI en 2026-05-19T00:00:00Z arXiv:2605.16090: CrossMPI — an attack on vision-language models using image-only perturbation https://24-ai.news/en/news/2026-05-19/arxiv-forge-agent-memory/ 24 AI en 2026-05-19T00:00:00Z arXiv:2605.16233: FORGE — AI agents develop shared memory without fine-tuning https://24-ai.news/en/news/2026-05-19/arxiv-llm-confabulation-scaling-law/ 24 AI en 2026-05-19T00:00:00Z arXiv:2605.18732: Scaling Law for Hallucinations — Larger Model Does Not Always Mean Fewer Errors https://24-ai.news/en/news/2026-05-19/arxiv-llm-tree-disease-forecasting/ 24 AI en 2026-05-19T00:00:00Z arXiv:2605.16238: LLM-guided tree search beats CDC in epidemic forecasting https://24-ai.news/en/news/2026-05-19/arxiv-prompt-injection-contextual-integrity/ 24 AI en 2026-05-19T00:00:00Z arXiv:2605.17634: Why Data/Instruction Separation Cannot Stop Prompt Injection https://24-ai.news/en/news/2026-05-19/arxiv-rope-positional-encoding-limits/ 24 AI en 2026-05-19T00:00:00Z arXiv:2605.15514: RoPE mathematically cannot distinguish positions or tokens in long contexts — theoretical proof of a fundamental limitation https://24-ai.news/en/news/2026-05-19/claude-api-web-search-sec/ 24 AI en 2026-05-19T00:00:00Z Anthropic: Claude API web search tool now returns enriched data from SEC filings https://24-ai.news/en/news/2026-05-19/claude-code-v2-1-144-bg-sessions/ 24 AI en 2026-05-19T00:00:00Z Anthropic Claude Code: v2.1.144 Brings /resume for Background Sessions and Fix for 75-Second Hang https://24-ai.news/en/news/2026-05-19/eu-ai-office-high-risk-consultation/ 24 AI en 2026-05-19T00:00:00Z EU AI Office: Public Consultation on High-Risk AI System Classification — Deadline June 23, 2026 https://24-ai.news/en/news/2026-05-19/github-copilot-remote-control-ga/ 24 AI en 2026-05-19T00:00:00Z GitHub: Copilot CLI remote control now generally available on all platforms https://24-ai.news/en/news/2026-05-19/github-copilot-spaces-api-ga/ 24 AI en 2026-05-19T00:00:00Z GitHub: Copilot Spaces API now generally available https://24-ai.news/en/news/2026-05-19/kubectl-debug-evidence-gap/ 24 AI en 2026-05-19T00:00:00Z CNCF: Kubernetes debugger erases traces — a serious problem for security audits https://24-ai.news/en/news/2026-05-19/pytorch-executorch-mlx-apple/ 24 AI en 2026-05-19T00:00:00Z PyTorch: ExecuTorch MLX Delegate delivers 3–6× faster model execution on Apple Silicon GPUs https://24-ai.news/de/news/2026-05-19/anthropic-acquires-stainless-sdk/ 24 AI de 2026-05-19T00:00:00Z Anthropic: Übernahme von Stainless integriert MCP-Server-Tooling und SDK-Entwicklung direkt in die Claude-Plattform https://24-ai.news/de/news/2026-05-19/anthropic-mcp-tunnels-sandboxes/ 24 AI de 2026-05-19T00:00:00Z Anthropic: MCP Tunnels, Self-Hosted Sandboxes und automatisches File-Spill für Agenten https://24-ai.news/de/news/2026-05-19/arxiv-ai-auto-research-roadmap/ 24 AI de 2026-05-19T00:00:00Z arXiv:2605.18661: KI für automatisierte Forschung — Roadmap und Anwenderhandbuch https://24-ai.news/de/news/2026-05-19/arxiv-code-as-agent-harness/ 24 AI de 2026-05-19T00:00:00Z arXiv:2605.18747: Code als operatives Substrat — ein neues KI-Agenten-Paradigma https://24-ai.news/de/news/2026-05-19/arxiv-crossmpi-vlm-injection/ 24 AI de 2026-05-19T00:00:00Z arXiv:2605.16090: CrossMPI — Angriff auf Vision-Language-Modelle ausschließlich durch Bildperturbation https://24-ai.news/de/news/2026-05-19/arxiv-forge-agent-memory/ 24 AI de 2026-05-19T00:00:00Z arXiv:2605.16233: FORGE — KI-Agenten entwickeln gemeinsames Gedächtnis ohne Fine-Tuning https://24-ai.news/de/news/2026-05-19/arxiv-llm-confabulation-scaling-law/ 24 AI de 2026-05-19T00:00:00Z arXiv:2605.18732: Skalierungsgesetz für Halluzinationen — größeres Modell bedeutet nicht zwingend weniger Fehler https://24-ai.news/de/news/2026-05-19/arxiv-llm-tree-disease-forecasting/ 24 AI de 2026-05-19T00:00:00Z arXiv:2605.16238: LLM-geführte Baumsuche übertrifft CDC bei Epidemieprognosen https://24-ai.news/de/news/2026-05-19/arxiv-prompt-injection-contextual-integrity/ 24 AI de 2026-05-19T00:00:00Z arXiv:2605.17634: Warum Data/Instruction Separation Prompt Injection nicht stoppen kann https://24-ai.news/de/news/2026-05-19/arxiv-rope-positional-encoding-limits/ 24 AI de 2026-05-19T00:00:00Z arXiv:2605.15514: RoPE kann in langen Kontexten weder Positionen noch Token unterscheiden — theoretischer Beweis einer fundamentalen Einschränkung https://24-ai.news/de/news/2026-05-19/claude-api-web-search-sec/ 24 AI de 2026-05-19T00:00:00Z Anthropic: Claude-API-Websuchtools liefert jetzt strukturierte Daten aus SEC-Einreichungen https://24-ai.news/de/news/2026-05-19/claude-code-v2-1-144-bg-sessions/ 24 AI de 2026-05-19T00:00:00Z Anthropic Claude Code: v2.1.144 bringt /resume für Background-Sitzungen und Fix für 75-Sekunden-Hang https://24-ai.news/de/news/2026-05-19/eu-ai-office-high-risk-consultation/ 24 AI de 2026-05-19T00:00:00Z EU AI Office: Öffentliche Konsultation zur Klassifizierung von Hochrisiko-KI-Systemen — Frist 23. Juni 2026 https://24-ai.news/de/news/2026-05-19/github-copilot-remote-control-ga/ 24 AI de 2026-05-19T00:00:00Z GitHub: Copilot-CLI-Fernsteuerung jetzt allgemein verfügbar auf allen Plattformen https://24-ai.news/de/news/2026-05-19/github-copilot-spaces-api-ga/ 24 AI de 2026-05-19T00:00:00Z GitHub: Copilot Spaces API jetzt allgemein verfügbar https://24-ai.news/de/news/2026-05-19/kubectl-debug-evidence-gap/ 24 AI de 2026-05-19T00:00:00Z CNCF: Kubernetes-Debugger löscht Spuren — ernstes Problem für Sicherheitsaudits https://24-ai.news/de/news/2026-05-19/pytorch-executorch-mlx-apple/ 24 AI de 2026-05-19T00:00:00Z PyTorch: ExecuTorch MLX Delegate bringt 3–6× schnellere Modellausführung auf Apple-Silicon-GPUs https://24-ai.news/zh/news/2026-05-19/anthropic-acquires-stainless-sdk/ 24 AI zh 2026-05-19T00:00:00Z Anthropic: 收购Stainless，将MCP服务器工具与SDK开发直接整合至Claude平台 https://24-ai.news/zh/news/2026-05-19/anthropic-mcp-tunnels-sandboxes/ 24 AI zh 2026-05-19T00:00:00Z Anthropic：MCP Tunnels、自托管沙盒与工具输出自动文件溢出 https://24-ai.news/zh/news/2026-05-19/arxiv-ai-auto-research-roadmap/ 24 AI zh 2026-05-19T00:00:00Z arXiv:2605.18661：AI自动研究路线图与用户指南 https://24-ai.news/zh/news/2026-05-19/arxiv-code-as-agent-harness/ 24 AI zh 2026-05-19T00:00:00Z arXiv:2605.18747：代码作为操作底层——AI代理的新范式 https://24-ai.news/zh/news/2026-05-19/arxiv-crossmpi-vlm-injection/ 24 AI zh 2026-05-19T00:00:00Z arXiv:2605.16090: CrossMPI——仅通过图像扰动对视觉语言模型发起的跨模态攻击 https://24-ai.news/zh/news/2026-05-19/arxiv-forge-agent-memory/ 24 AI zh 2026-05-19T00:00:00Z arXiv:2605.16233: FORGE——AI智能体无需微调即可构建共享记忆 https://24-ai.news/zh/news/2026-05-19/arxiv-llm-confabulation-scaling-law/ 24 AI zh 2026-05-19T00:00:00Z arXiv:2605.18732：幻觉的缩放定律——更大的模型并不一定意味着更少的错误 https://24-ai.news/zh/news/2026-05-19/arxiv-llm-tree-disease-forecasting/ 24 AI zh 2026-05-19T00:00:00Z arXiv:2605.16238: LLM引导树搜索系统在疾病流行预测中超越CDC专家模型 https://24-ai.news/zh/news/2026-05-19/arxiv-prompt-injection-contextual-integrity/ 24 AI zh 2026-05-19T00:00:00Z arXiv:2605.17634：为何数据与指令分离无法阻止提示注入攻击？ https://24-ai.news/zh/news/2026-05-19/arxiv-rope-positional-encoding-limits/ 24 AI zh 2026-05-19T00:00:00Z arXiv:2605.15514: RoPE在长上下文中无法区分位置或词元——基本限制的理论证明 https://24-ai.news/zh/news/2026-05-19/claude-api-web-search-sec/ 24 AI zh 2026-05-19T00:00:00Z Anthropic: Claude API网络搜索工具现可返回来自SEC申报文件的富结构化数据 https://24-ai.news/zh/news/2026-05-19/claude-code-v2-1-144-bg-sessions/ 24 AI zh 2026-05-19T00:00:00Z Anthropic Claude Code v2.1.144：/resume支持后台会话与修复75秒挂起问题 https://24-ai.news/zh/news/2026-05-19/eu-ai-office-high-risk-consultation/ 24 AI zh 2026-05-19T00:00:00Z EU AI Office：高风险AI系统分类公开征询意见——截止日期2026年6月23日 https://24-ai.news/zh/news/2026-05-19/github-copilot-remote-control-ga/ 24 AI zh 2026-05-19T00:00:00Z GitHub: Copilot CLI远程控制功能现已在所有平台全面开放 https://24-ai.news/zh/news/2026-05-19/github-copilot-spaces-api-ga/ 24 AI zh 2026-05-19T00:00:00Z GitHub: Copilot Spaces API现已全面开放 https://24-ai.news/zh/news/2026-05-19/kubectl-debug-evidence-gap/ 24 AI zh 2026-05-19T00:00:00Z CNCF: Kubernetes调试器删除访问痕迹——安全审计面临严重合规问题 https://24-ai.news/zh/news/2026-05-19/pytorch-executorch-mlx-apple/ 24 AI zh 2026-05-19T00:00:00Z PyTorch: ExecuTorch MLX Delegate在Apple Silicon GPU上实现3–6倍模型推理加速 https://24-ai.news/ja/news/2026-05-19/anthropic-acquires-stainless-sdk/ 24 AI ja 2026-05-19T00:00:00Z Anthropic: StainlessのM&Aにより、MCPサーバーツールとSDK開発をClaudeプラットフォームに直接統合 https://24-ai.news/ja/news/2026-05-19/anthropic-mcp-tunnels-sandboxes/ 24 AI ja 2026-05-19T00:00:00Z Anthropic：MCP Tunnels、セルフホストサンドボックス、ツール出力の自動ファイルスピル https://24-ai.news/ja/news/2026-05-19/arxiv-ai-auto-research-roadmap/ 24 AI ja 2026-05-19T00:00:00Z arXiv:2605.18661：AI自動研究のロードマップとユーザーガイド https://24-ai.news/ja/news/2026-05-19/arxiv-code-as-agent-harness/ 24 AI ja 2026-05-19T00:00:00Z arXiv:2605.18747：操作基盤としてのコード——AIエージェントの新しいパラダイム https://24-ai.news/ja/news/2026-05-19/arxiv-crossmpi-vlm-injection/ 24 AI ja 2026-05-19T00:00:00Z arXiv:2605.16090: CrossMPI――画像の摂動のみで視覚言語モデルを攻撃するクロスモーダルプロンプトインジェクション https://24-ai.news/ja/news/2026-05-19/arxiv-forge-agent-memory/ 24 AI ja 2026-05-19T00:00:00Z arXiv:2605.16233: FORGE――ファインチューニングなしでAIエージェントが共有メモリを獲得する手法 https://24-ai.news/ja/news/2026-05-19/arxiv-llm-confabulation-scaling-law/ 24 AI ja 2026-05-19T00:00:00Z arXiv:2605.18732：幻覚のスケーリング則——大きなモデルが必ずしも少ないエラーを意味しない https://24-ai.news/ja/news/2026-05-19/arxiv-llm-tree-disease-forecasting/ 24 AI ja 2026-05-19T00:00:00Z arXiv:2605.16238: LLM誘導ツリーサーチが疾患流行予測でCDCを超える精度を実現 https://24-ai.news/ja/news/2026-05-19/arxiv-prompt-injection-contextual-integrity/ 24 AI ja 2026-05-19T00:00:00Z arXiv:2605.17634：データと命令の分離ではプロンプトインジェクションを防げない理由 https://24-ai.news/ja/news/2026-05-19/arxiv-rope-positional-encoding-limits/ 24 AI ja 2026-05-19T00:00:00Z arXiv:2605.15514: RoPEは長いコンテキストで位置もトークンも区別できない――基本的限界の理論的証明 https://24-ai.news/ja/news/2026-05-19/claude-api-web-search-sec/ 24 AI ja 2026-05-19T00:00:00Z Anthropic: Claude APIのウェブ検索ツールがSEC申告書の充実した構造化データを返すよう更新 https://24-ai.news/ja/news/2026-05-19/claude-code-v2-1-144-bg-sessions/ 24 AI ja 2026-05-19T00:00:00Z Anthropic Claude Code v2.1.144：バックグラウンドセッションへの/resumeサポートと75秒ハングの修正 https://24-ai.news/ja/news/2026-05-19/eu-ai-office-high-risk-consultation/ 24 AI ja 2026-05-19T00:00:00Z EU AI Office：高リスクAIシステム分類に関するパブリックコンサルテーション——期限は2026年6月23日 https://24-ai.news/ja/news/2026-05-19/github-copilot-remote-control-ga/ 24 AI ja 2026-05-19T00:00:00Z GitHub: Copilot CLIのリモートコントロールがすべてのプラットフォームで一般提供開始 https://24-ai.news/ja/news/2026-05-19/github-copilot-spaces-api-ga/ 24 AI ja 2026-05-19T00:00:00Z GitHub: Copilot Spaces APIが一般提供開始 https://24-ai.news/ja/news/2026-05-19/kubectl-debug-evidence-gap/ 24 AI ja 2026-05-19T00:00:00Z CNCF: Kubernetesデバッガがアクセス痕跡を削除――セキュリティ監査に深刻な問題 https://24-ai.news/ja/news/2026-05-19/pytorch-executorch-mlx-apple/ 24 AI ja 2026-05-19T00:00:00Z PyTorch: ExecuTorch MLX DelegateがApple Silicon GPUでモデル推論を3〜6倍高速化 https://24-ai.news/ko/news/2026-05-19/anthropic-acquires-stainless-sdk/ 24 AI ko 2026-05-19T00:00:00Z Anthropic: Stainless 인수로 MCP 서버 툴링과 SDK 개발을 Claude 플랫폼에 직접 통합 https://24-ai.news/ko/news/2026-05-19/anthropic-mcp-tunnels-sandboxes/ 24 AI ko 2026-05-19T00:00:00Z Anthropic: MCP Tunnels, 자체 호스팅 샌드박스, 툴 출력 자동 파일 스필 https://24-ai.news/ko/news/2026-05-19/arxiv-ai-auto-research-roadmap/ 24 AI ko 2026-05-19T00:00:00Z arXiv:2605.18661: AI 자동 연구 로드맵 및 사용자 가이드 https://24-ai.news/ko/news/2026-05-19/arxiv-code-as-agent-harness/ 24 AI ko 2026-05-19T00:00:00Z arXiv:2605.18747: 운영 기반으로서의 코드——AI 에이전트의 새로운 패러다임 https://24-ai.news/ko/news/2026-05-19/arxiv-crossmpi-vlm-injection/ 24 AI ko 2026-05-19T00:00:00Z arXiv:2605.16090: CrossMPI――이미지 섭동만으로 비전-언어 모델을 공격하는 크로스모달 프롬프트 인젝션 https://24-ai.news/ko/news/2026-05-19/arxiv-forge-agent-memory/ 24 AI ko 2026-05-19T00:00:00Z arXiv:2605.16233: FORGE――파인튜닝 없이 AI 에이전트가 공유 메모리를 구축하는 방법 https://24-ai.news/ko/news/2026-05-19/arxiv-llm-confabulation-scaling-law/ 24 AI ko 2026-05-19T00:00:00Z arXiv:2605.18732: 환각의 스케일링 법칙——더 큰 모델이 반드시 더 적은 오류를 의미하지는 않습니다 https://24-ai.news/ko/news/2026-05-19/arxiv-llm-tree-disease-forecasting/ 24 AI ko 2026-05-19T00:00:00Z arXiv:2605.16238: LLM 유도 트리 서치가 질병 유행 예측에서 CDC를 능가하다 https://24-ai.news/ko/news/2026-05-19/arxiv-prompt-injection-contextual-integrity/ 24 AI ko 2026-05-19T00:00:00Z arXiv:2605.17634: 데이터와 명령어 분리로는 프롬프트 인젝션을 막을 수 없는 이유 https://24-ai.news/ko/news/2026-05-19/arxiv-rope-positional-encoding-limits/ 24 AI ko 2026-05-19T00:00:00Z arXiv:2605.15514: RoPE는 긴 컨텍스트에서 위치도 토큰도 구별할 수 없다――근본적 한계의 이론적 증명 https://24-ai.news/ko/news/2026-05-19/claude-api-web-search-sec/ 24 AI ko 2026-05-19T00:00:00Z Anthropic: Claude API 웹 검색 도구가 SEC 신고서에서 풍부한 구조화 데이터를 반환하도록 업데이트 https://24-ai.news/ko/news/2026-05-19/claude-code-v2-1-144-bg-sessions/ 24 AI ko 2026-05-19T00:00:00Z Anthropic Claude Code v2.1.144: 백그라운드 세션 /resume 지원 및 75초 행 수정 https://24-ai.news/ko/news/2026-05-19/eu-ai-office-high-risk-consultation/ 24 AI ko 2026-05-19T00:00:00Z EU AI Office: 고위험 AI 시스템 분류에 관한 공개 의견 수렴——마감 2026년 6월 23일 https://24-ai.news/ko/news/2026-05-19/github-copilot-remote-control-ga/ 24 AI ko 2026-05-19T00:00:00Z GitHub: Copilot CLI 원격 제어 기능이 모든 플랫폼에서 일반 제공 개시 https://24-ai.news/ko/news/2026-05-19/github-copilot-spaces-api-ga/ 24 AI ko 2026-05-19T00:00:00Z GitHub: Copilot Spaces API 일반 제공 개시 https://24-ai.news/ko/news/2026-05-19/kubectl-debug-evidence-gap/ 24 AI ko 2026-05-19T00:00:00Z CNCF: Kubernetes 디버거가 접근 흔적을 삭제――보안 감사에 심각한 규정 준수 문제 https://24-ai.news/ko/news/2026-05-19/pytorch-executorch-mlx-apple/ 24 AI ko 2026-05-19T00:00:00Z PyTorch: ExecuTorch MLX Delegate가 Apple Silicon GPU에서 모델 추론을 3~6배 가속 https://24-ai.news/hr/news/2026-05-18/arxiv-argus-evidence-assembly-deep-research/ 24 AI hr 2026-05-18T00:00:00Z arXiv:2605.16217 Argus: evidence assembly arhitektura za deep research agente postiže +12,7pp s 8 paralelnih searchera https://24-ai.news/hr/news/2026-05-18/arxiv-case-based-calibration-llm-tool-use/ 24 AI hr 2026-05-18T00:00:00Z arXiv:2605.15041 CAST framework: case-based calibration LLM tool use postiže +5,85pp BFCLv2 i -26 % reasoning length https://24-ai.news/hr/news/2026-05-18/arxiv-differentiable-mixture-of-agents-swarm/ 24 AI hr 2026-05-18T00:00:00Z arXiv:2605.15706 Differentiable Mixture-of-Agents: dinamičko routing aktiviranje agenata postiže SOTA kroz 9 benchmarka https://24-ai.news/hr/news/2026-05-18/arxiv-dual-dimensional-consistency-adaptive-inference/ 24 AI hr 2026-05-18T00:00:00Z arXiv:2605.15100 Dual-Dimensional Consistency: 10× redukcija token consumption uz održanu accuracy kroz pet benchmarka https://24-ai.news/hr/news/2026-05-18/arxiv-sleeper-memory-poisoning-llm-agents/ 24 AI hr 2026-05-18T00:00:00Z arXiv:2605.15338 Sleeper Memory Poisoning: 99,8 % attack success rate na GPT-5.5 kroz perzistentnu memoriju LLM agenata https://24-ai.news/hr/news/2026-05-18/arxiv-small-private-lm-educational-assessment/ 24 AI hr 2026-05-18T00:00:00Z arXiv:2605.15015 Small Private LM: kompetitivni rezultati u educational assessment design uz human-in-the-loop preporuke https://24-ai.news/hr/news/2026-05-18/arxiv-traversal-context-agentic-graphrag/ 24 AI hr 2026-05-18T00:00:00Z arXiv:2605.15109 Traversal Context: agentic GraphRAG mora dokumentirati visited-but-uncited entitete za pravu provenance https://24-ai.news/hr/news/2026-05-18/databricks-veeva-vault-crm-ai-agents/ 24 AI hr 2026-05-18T00:00:00Z Databricks + Veeva Vault CRM: tri specijalizirana AI agenta za life sciences commercial workflow https://24-ai.news/hr/news/2026-05-18/github-copilot-gpt-5-3-codex-business-enterprise-base/ 24 AI hr 2026-05-18T00:00:00Z GitHub Copilot: GPT-5.3-Codex postaje base model za Business i Enterprise s 12-mjesečnim LTS jamstvom https://24-ai.news/hr/news/2026-05-18/github-grok-code-fast-1-deprecated/ 24 AI hr 2026-05-18T00:00:00Z GitHub Copilot: Grok Code Fast 1 deprekiran 15. svibnja 2026.; preporučene zamjene GPT-5 mini i Claude Haiku 4.5 https://24-ai.news/hr/news/2026-05-18/xai-sdk-python-1-13-0-batch-video-extension/ 24 AI hr 2026-05-18T00:00:00Z xAI SDK Python v1.13.0: prepare_extension() omogućuje batch video extension za serije generiranih klipova https://24-ai.news/en/news/2026-05-18/arxiv-argus-evidence-assembly-deep-research/ 24 AI en 2026-05-18T00:00:00Z arXiv:2605.16217 Argus: evidence assembly architecture for deep research agents achieves +12.7pp with 8 parallel searchers https://24-ai.news/en/news/2026-05-18/arxiv-case-based-calibration-llm-tool-use/ 24 AI en 2026-05-18T00:00:00Z arXiv:2605.15041 CAST Framework: Case-Based Calibration for LLM Tool Use Achieves +5.85pp BFCLv2 and -26% Reasoning Length https://24-ai.news/en/news/2026-05-18/arxiv-differentiable-mixture-of-agents-swarm/ 24 AI en 2026-05-18T00:00:00Z arXiv:2605.15706 Differentiable Mixture-of-Agents: dynamic per-step agent routing achieves SOTA across 9 benchmarks https://24-ai.news/en/news/2026-05-18/arxiv-dual-dimensional-consistency-adaptive-inference/ 24 AI en 2026-05-18T00:00:00Z arXiv:2605.15100 Dual-Dimensional Consistency: 10× Token Consumption Reduction with Maintained Accuracy Across Five Benchmarks https://24-ai.news/en/news/2026-05-18/arxiv-sleeper-memory-poisoning-llm-agents/ 24 AI en 2026-05-18T00:00:00Z arXiv:2605.15338 Sleeper Memory Poisoning: 99.8% attack success rate on GPT-5.5 via persistent memory of LLM agents https://24-ai.news/en/news/2026-05-18/arxiv-small-private-lm-educational-assessment/ 24 AI en 2026-05-18T00:00:00Z arXiv:2605.15015 Small Private LM: Competitive Results in Educational Assessment Design with Human-in-the-Loop Recommendations https://24-ai.news/en/news/2026-05-18/arxiv-traversal-context-agentic-graphrag/ 24 AI en 2026-05-18T00:00:00Z arXiv:2605.15109 Traversal Context: Agentic GraphRAG Must Document Visited-but-Uncited Entities for True Provenance https://24-ai.news/en/news/2026-05-18/databricks-veeva-vault-crm-ai-agents/ 24 AI en 2026-05-18T00:00:00Z Databricks + Veeva Vault CRM: three specialized AI agents for life sciences commercial workflows https://24-ai.news/en/news/2026-05-18/github-copilot-gpt-5-3-codex-business-enterprise-base/ 24 AI en 2026-05-18T00:00:00Z GitHub Copilot: GPT-5.3-Codex becomes base model for Business and Enterprise with 12-month LTS guarantee https://24-ai.news/en/news/2026-05-18/github-grok-code-fast-1-deprecated/ 24 AI en 2026-05-18T00:00:00Z GitHub Copilot: Grok Code Fast 1 Deprecated May 15, 2026; Recommended Replacements GPT-5 mini and Claude Haiku 4.5 https://24-ai.news/en/news/2026-05-18/xai-sdk-python-1-13-0-batch-video-extension/ 24 AI en 2026-05-18T00:00:00Z xAI SDK Python v1.13.0: prepare_extension() Enables Batch Video Extension for Generated Clip Series https://24-ai.news/de/news/2026-05-18/arxiv-argus-evidence-assembly-deep-research/ 24 AI de 2026-05-18T00:00:00Z arXiv:2605.16217 Argus: Evidence-Assembly-Architektur für Deep-Research-Agenten erreicht +12,7 Pkt. mit 8 parallelen Suchern https://24-ai.news/de/news/2026-05-18/arxiv-case-based-calibration-llm-tool-use/ 24 AI de 2026-05-18T00:00:00Z arXiv:2605.15041 CAST-Framework: Case-Based Calibration für LLM Tool Use erreicht +5,85pp BFCLv2 und -26 % Reasoning-Länge https://24-ai.news/de/news/2026-05-18/arxiv-differentiable-mixture-of-agents-swarm/ 24 AI de 2026-05-18T00:00:00Z arXiv:2605.15706 Differentiable Mixture-of-Agents: dynamisches Per-Step-Agenten-Routing erreicht SOTA auf 9 Benchmarks https://24-ai.news/de/news/2026-05-18/arxiv-dual-dimensional-consistency-adaptive-inference/ 24 AI de 2026-05-18T00:00:00Z arXiv:2605.15100 Dual-Dimensional Consistency: 10-fache Reduzierung des Token-Verbrauchs bei gleichbleibender Genauigkeit über fünf Benchmarks https://24-ai.news/de/news/2026-05-18/arxiv-sleeper-memory-poisoning-llm-agents/ 24 AI de 2026-05-18T00:00:00Z arXiv:2605.15338 Sleeper Memory Poisoning: 99,8 % Angriffserfolgsrate auf GPT-5.5 über persistenten Speicher von LLM-Agenten https://24-ai.news/de/news/2026-05-18/arxiv-small-private-lm-educational-assessment/ 24 AI de 2026-05-18T00:00:00Z arXiv:2605.15015 Small Private LM: Wettbewerbsfähige Ergebnisse im Bildungsassessment-Design mit Human-in-the-Loop-Empfehlungen https://24-ai.news/de/news/2026-05-18/arxiv-traversal-context-agentic-graphrag/ 24 AI de 2026-05-18T00:00:00Z arXiv:2605.15109 Traversal Context: Agentisches GraphRAG muss besuchte, aber nicht zitierte Entitäten für echte Provenienz dokumentieren https://24-ai.news/de/news/2026-05-18/databricks-veeva-vault-crm-ai-agents/ 24 AI de 2026-05-18T00:00:00Z Databricks + Veeva Vault CRM: drei spezialisierte KI-Agenten für Life-Sciences-Commercial-Workflows https://24-ai.news/de/news/2026-05-18/github-copilot-gpt-5-3-codex-business-enterprise-base/ 24 AI de 2026-05-18T00:00:00Z GitHub Copilot: GPT-5.3-Codex wird Basismodell für Business und Enterprise mit 12-monatiger LTS-Garantie https://24-ai.news/de/news/2026-05-18/github-grok-code-fast-1-deprecated/ 24 AI de 2026-05-18T00:00:00Z GitHub Copilot: Grok Code Fast 1 am 15. Mai 2026 abgekündigt; empfohlene Alternativen GPT-5 mini und Claude Haiku 4.5 https://24-ai.news/de/news/2026-05-18/xai-sdk-python-1-13-0-batch-video-extension/ 24 AI de 2026-05-18T00:00:00Z xAI SDK Python v1.13.0: prepare_extension() ermöglicht Batch-Video-Extension für generierte Clip-Serien https://24-ai.news/zh/news/2026-05-18/arxiv-argus-evidence-assembly-deep-research/ 24 AI zh 2026-05-18T00:00:00Z arXiv:2605.16217 Argus：深度研究智能体的证据汇编架构，8路并行搜索器提升12.7个百分点 https://24-ai.news/zh/news/2026-05-18/arxiv-case-based-calibration-llm-tool-use/ 24 AI zh 2026-05-18T00:00:00Z arXiv:2605.15041 CAST 框架：基于案例的 LLM 工具使用校准在 BFCLv2 上提升5.85个百分点，推理长度减少26% https://24-ai.news/zh/news/2026-05-18/arxiv-differentiable-mixture-of-agents-swarm/ 24 AI zh 2026-05-18T00:00:00Z arXiv:2605.15706 可微分智能体混合：动态路由智能体激活在9个基准测试中达SOTA https://24-ai.news/zh/news/2026-05-18/arxiv-dual-dimensional-consistency-adaptive-inference/ 24 AI zh 2026-05-18T00:00:00Z arXiv:2605.15100 双维度一致性：在五项基准测试中将token消耗减少10倍同时保持准确性 https://24-ai.news/zh/news/2026-05-18/arxiv-sleeper-memory-poisoning-llm-agents/ 24 AI zh 2026-05-18T00:00:00Z arXiv:2605.15338 休眠记忆投毒：通过LLM智能体持久化记忆对GPT-5.5的攻击成功率达99.8% https://24-ai.news/zh/news/2026-05-18/arxiv-small-private-lm-educational-assessment/ 24 AI zh 2026-05-18T00:00:00Z arXiv:2605.15015 小型私有语言模型：在教育评估设计中取得有竞争力的结果，并建议采用人机协作方式 https://24-ai.news/zh/news/2026-05-18/arxiv-traversal-context-agentic-graphrag/ 24 AI zh 2026-05-18T00:00:00Z arXiv:2605.15109 遍历上下文：智能体 GraphRAG 必须记录已访问但未引用的实体以实现真正的溯源 https://24-ai.news/zh/news/2026-05-18/databricks-veeva-vault-crm-ai-agents/ 24 AI zh 2026-05-18T00:00:00Z Databricks + Veeva Vault CRM：面向生命科学商业工作流的三款专业AI智能体 https://24-ai.news/zh/news/2026-05-18/github-copilot-gpt-5-3-codex-business-enterprise-base/ 24 AI zh 2026-05-18T00:00:00Z GitHub Copilot：GPT-5.3-Codex成为Business和Enterprise基础模型，享有12个月LTS保障 https://24-ai.news/zh/news/2026-05-18/github-grok-code-fast-1-deprecated/ 24 AI zh 2026-05-18T00:00:00Z GitHub Copilot：Grok Code Fast 1 于2026年5月15日弃用；推荐替代方案为 GPT-5 mini 和 Claude Haiku 4.5 https://24-ai.news/zh/news/2026-05-18/xai-sdk-python-1-13-0-batch-video-extension/ 24 AI zh 2026-05-18T00:00:00Z xAI SDK Python v1.13.0：prepare_extension() 为系列生成视频片段提供批量视频扩展支持 https://24-ai.news/ja/news/2026-05-18/arxiv-argus-evidence-assembly-deep-research/ 24 AI ja 2026-05-18T00:00:00Z arXiv:2605.16217 Argus：深層調査エージェント向けエビデンス・アセンブリ・アーキテクチャ、8並列サーチャーで+12.7pp達成 https://24-ai.news/ja/news/2026-05-18/arxiv-case-based-calibration-llm-tool-use/ 24 AI ja 2026-05-18T00:00:00Z arXiv:2605.15041 CAST フレームワーク：事例ベースキャリブレーション LLM ツール使用が BFCLv2 で +5.85pp、推論長を26%削減 https://24-ai.news/ja/news/2026-05-18/arxiv-differentiable-mixture-of-agents-swarm/ 24 AI ja 2026-05-18T00:00:00Z arXiv:2605.15706 微分可能エージェント混合：動的ルーティングによるエージェント起動で9ベンチマーク中SOTA達成 https://24-ai.news/ja/news/2026-05-18/arxiv-dual-dimensional-consistency-adaptive-inference/ 24 AI ja 2026-05-18T00:00:00Z arXiv:2605.15100 デュアルディメンション一貫性：5つのベンチマークで精度を維持しながらトークン消費量を10倍以上削減 https://24-ai.news/ja/news/2026-05-18/arxiv-sleeper-memory-poisoning-llm-agents/ 24 AI ja 2026-05-18T00:00:00Z arXiv:2605.15338 スリーパー・メモリ・ポイズニング：LLMエージェントの永続メモリを介したGPT-5.5への攻撃成功率99.8% https://24-ai.news/ja/news/2026-05-18/arxiv-small-private-lm-educational-assessment/ 24 AI ja 2026-05-18T00:00:00Z arXiv:2605.15015 Small Private LM：教育評価設計において競争力のある結果を達成し、ヒューマン・イン・ザ・ループの導入を推奨 https://24-ai.news/ja/news/2026-05-18/arxiv-traversal-context-agentic-graphrag/ 24 AI ja 2026-05-18T00:00:00Z arXiv:2605.15109 トラバーサルコンテキスト：エージェント型 GraphRAG は真のプロベナンスのために訪問済み・未引用エンティティを記録しなければならない https://24-ai.news/ja/news/2026-05-18/databricks-veeva-vault-crm-ai-agents/ 24 AI ja 2026-05-18T00:00:00Z Databricks + Veeva Vault CRM：ライフサイエンス商業ワークフロー向け3種の専門AIエージェント https://24-ai.news/ja/news/2026-05-18/github-copilot-gpt-5-3-codex-business-enterprise-base/ 24 AI ja 2026-05-18T00:00:00Z GitHub Copilot：GPT-5.3-CodexがBusinessおよびEnterpriseの基盤モデルに、12ヶ月LTS保証付き https://24-ai.news/ja/news/2026-05-18/github-grok-code-fast-1-deprecated/ 24 AI ja 2026-05-18T00:00:00Z GitHub Copilot：Grok Code Fast 1 が2026年5月15日に非推奨化；推奨代替モデルは GPT-5 mini と Claude Haiku 4.5 https://24-ai.news/ja/news/2026-05-18/xai-sdk-python-1-13-0-batch-video-extension/ 24 AI ja 2026-05-18T00:00:00Z xAI SDK Python v1.13.0：prepare_extension() により生成クリップシリーズのバッチビデオ拡張が可能に https://24-ai.news/ko/news/2026-05-18/arxiv-argus-evidence-assembly-deep-research/ 24 AI ko 2026-05-18T00:00:00Z arXiv:2605.16217 Argus: 심층 연구 에이전트를 위한 증거 조립 아키텍처, 8개 병렬 검색기로 +12.7pp 달성 https://24-ai.news/ko/news/2026-05-18/arxiv-case-based-calibration-llm-tool-use/ 24 AI ko 2026-05-18T00:00:00Z arXiv:2605.15041 CAST 프레임워크: 사례 기반 보정 LLM 도구 사용으로 BFCLv2 +5.85pp 달성 및 추론 길이 26% 감소 https://24-ai.news/ko/news/2026-05-18/arxiv-differentiable-mixture-of-agents-swarm/ 24 AI ko 2026-05-18T00:00:00Z arXiv:2605.15706 미분 가능한 에이전트 혼합: 동적 라우팅 에이전트 활성화로 9개 벤치마크 SOTA 달성 https://24-ai.news/ko/news/2026-05-18/arxiv-dual-dimensional-consistency-adaptive-inference/ 24 AI ko 2026-05-18T00:00:00Z arXiv:2605.15100 이중 차원 일관성: 다섯 가지 벤치마크에서 정확도를 유지하면서 토큰 소비 10배 이상 감소 https://24-ai.news/ko/news/2026-05-18/arxiv-sleeper-memory-poisoning-llm-agents/ 24 AI ko 2026-05-18T00:00:00Z arXiv:2605.15338 슬리퍼 메모리 포이즈닝: LLM 에이전트 영구 메모리를 통한 GPT-5.5 공격 성공률 99.8% https://24-ai.news/ko/news/2026-05-18/arxiv-small-private-lm-educational-assessment/ 24 AI ko 2026-05-18T00:00:00Z arXiv:2605.15015 소형 프라이빗 LM: 교육 평가 설계에서 경쟁력 있는 결과 달성 및 휴먼 인 더 루프 권장 https://24-ai.news/ko/news/2026-05-18/arxiv-traversal-context-agentic-graphrag/ 24 AI ko 2026-05-18T00:00:00Z arXiv:2605.15109 순회 컨텍스트: 에이전트 GraphRAG는 진정한 출처 추적을 위해 방문했지만 인용하지 않은 엔티티를 기록해야 합니다 https://24-ai.news/ko/news/2026-05-18/databricks-veeva-vault-crm-ai-agents/ 24 AI ko 2026-05-18T00:00:00Z Databricks + Veeva Vault CRM: 생명과학 상업 워크플로를 위한 세 가지 전문 AI 에이전트 https://24-ai.news/ko/news/2026-05-18/github-copilot-gpt-5-3-codex-business-enterprise-base/ 24 AI ko 2026-05-18T00:00:00Z GitHub Copilot: GPT-5.3-Codex, Business 및 Enterprise 기본 모델로 전환, 12개월 LTS 보장 https://24-ai.news/ko/news/2026-05-18/github-grok-code-fast-1-deprecated/ 24 AI ko 2026-05-18T00:00:00Z GitHub Copilot: Grok Code Fast 1이 2026년 5월 15일 지원 중단; 권장 대안은 GPT-5 mini와 Claude Haiku 4.5 https://24-ai.news/ko/news/2026-05-18/xai-sdk-python-1-13-0-batch-video-extension/ 24 AI ko 2026-05-18T00:00:00Z xAI SDK Python v1.13.0: prepare_extension()으로 생성된 클립 시리즈의 배치 비디오 확장 지원