https://24-ai.news/hr/news/2026-07-08/anthropic-claude-code-2-1-205/ 24 AI hr 2026-07-08T00:00:00Z Claude Code v2.1.205: sigurnost sesija, popravci i ušteda od 400 MB RAM-a https://24-ai.news/hr/news/2026-07-08/anthropic-gram-dual-use-offswitch/ 24 AI hr 2026-07-08T00:00:00Z Anthropic razvija „off switch” za opasno znanje: GRAM izolira dual-use sposobnosti u uklonjive module https://24-ai.news/hr/news/2026-07-08/arxiv-doomed-early-abort/ 24 AI hr 2026-07-08T00:00:00Z Kaskada proba predviđa neuspjeh AI agenta iz skrivenih aktivacija pri prvom koraku i štedi do 47% tokena https://24-ai.news/hr/news/2026-07-08/arxiv-dt-guard-safety/ 24 AI hr 2026-07-08T00:00:00Z DT-Guard: model od 4B parametara nadmašuje 8B guardraile odvajanjem reasoning supervizije od brzine inferencije https://24-ai.news/hr/news/2026-07-08/arxiv-incontext-search-theory/ 24 AI hr 2026-07-08T00:00:00Z Formalna teorija precizno definira kada iterativna refleksija LLM-ova donosi eksponencijalna poboljšanja https://24-ai.news/hr/news/2026-07-08/aws-claude-apps-gateway/ 24 AI hr 2026-07-08T00:00:00Z AWS uvodi Claude Apps Gateway: centralno upravljanje Claude alatima za enterprise timove https://24-ai.news/hr/news/2026-07-08/cncf-cloud-native-ai-storage/ 24 AI hr 2026-07-08T00:00:00Z CNCF white paper: Pohrana podataka ostaje glavna prepreka cloud-native AI-ja u mjerilu https://24-ai.news/hr/news/2026-07-08/ibm-redhat-lightwell/ 24 AI hr 2026-07-08T00:00:00Z IBM i Red Hat proširuju Lightwell — 6.500+ saniranih open-source ovisnosti i clearinghouse za financije https://24-ai.news/hr/news/2026-07-08/langchain-nvidia-nemoclaw/ 24 AI hr 2026-07-08T00:00:00Z LangChain i NVIDIA lansiraju NemoClaw: otvoreni agentski stack postiže 10× niži trošak od konkurencije https://24-ai.news/hr/news/2026-07-08/microsoft-flint-viz-language/ 24 AI hr 2026-07-08T00:00:00Z Flint — Microsoftov open-source jezik koji kompajlira vizualizacije za AI agente https://24-ai.news/hr/news/2026-07-08/mistral-robostral-navigate/ 24 AI hr 2026-07-08T00:00:00Z Mistral Robostral Navigate: robotski AI koji navigira samo s RGB kamerom https://24-ai.news/hr/news/2026-07-08/openai-government-national-security/ 24 AI hr 2026-07-08T00:00:00Z OpenAI objavljuje načela za suradnju s vladama i sigurnosnim strukturama https://24-ai.news/hr/news/2026-07-08/openai-gpt-live/ 24 AI hr 2026-07-08T00:00:00Z OpenAI lansira GPT-Live: glasovni model za životne AI razgovore https://24-ai.news/hr/news/2026-07-08/openai-swe-bench-pro-critique/ 24 AI hr 2026-07-08T00:00:00Z OpenAI preispituje SWE-Bench Pro: vodeći benchmark za AI kodiranje ima probleme s pouzdanošću https://24-ai.news/hr/news/2026-07-08/pytorch-2-13-release/ 24 AI hr 2026-07-08T00:00:00Z PyTorch 2.13: do 4× manje GPU memorije za LLM treniranje i 12,3× brži FlexAttention https://24-ai.news/en/news/2026-07-08/anthropic-claude-code-2-1-205/ 24 AI en 2026-07-08T00:00:00Z Claude Code v2.1.205: Session Security, Bug Fixes, and 400 MB RAM Savings https://24-ai.news/en/news/2026-07-08/anthropic-gram-dual-use-offswitch/ 24 AI en 2026-07-08T00:00:00Z Anthropic Develops an 'Off Switch' for Dangerous Knowledge: GRAM Isolates Dual-Use Capabilities into Removable Modules https://24-ai.news/en/news/2026-07-08/arxiv-doomed-early-abort/ 24 AI en 2026-07-08T00:00:00Z A Probe Cascade Predicts AI Agent Failure from Hidden Activations at the First Step, Saving Up to 47% of Tokens https://24-ai.news/en/news/2026-07-08/arxiv-dt-guard-safety/ 24 AI en 2026-07-08T00:00:00Z DT-Guard: 4B-Parameter Model Outperforms 8B Guardrails by Separating Reasoning Supervision from Inference Speed https://24-ai.news/en/news/2026-07-08/arxiv-incontext-search-theory/ 24 AI en 2026-07-08T00:00:00Z Formal Theory Precisely Defines When Iterative LLM Reflection Yields Exponential Gains https://24-ai.news/en/news/2026-07-08/aws-claude-apps-gateway/ 24 AI en 2026-07-08T00:00:00Z AWS Introduces Claude Apps Gateway: Centralized Management of Claude Tools for Enterprise Teams https://24-ai.news/en/news/2026-07-08/cncf-cloud-native-ai-storage/ 24 AI en 2026-07-08T00:00:00Z CNCF White Paper: Data Storage Remains the Main Barrier to Cloud-Native AI at Scale https://24-ai.news/en/news/2026-07-08/ibm-redhat-lightwell/ 24 AI en 2026-07-08T00:00:00Z IBM and Red Hat Expand Lightwell — 6,500+ Remediated Open-Source Dependencies and Clearinghouse for Finance https://24-ai.news/en/news/2026-07-08/langchain-nvidia-nemoclaw/ 24 AI en 2026-07-08T00:00:00Z LangChain and NVIDIA Launch NemoClaw: Open Agent Stack Achieves 10× Lower Cost Than Competitors https://24-ai.news/en/news/2026-07-08/microsoft-flint-viz-language/ 24 AI en 2026-07-08T00:00:00Z Flint — Microsoft's Open-Source Language That Compiles Visualizations for AI Agents https://24-ai.news/en/news/2026-07-08/mistral-robostral-navigate/ 24 AI en 2026-07-08T00:00:00Z Mistral Robostral Navigate: Robotic AI That Navigates Using Only an RGB Camera https://24-ai.news/en/news/2026-07-08/openai-government-national-security/ 24 AI en 2026-07-08T00:00:00Z OpenAI Publishes Principles for Collaboration with Governments and Security Organizations https://24-ai.news/en/news/2026-07-08/openai-gpt-live/ 24 AI en 2026-07-08T00:00:00Z OpenAI Launches GPT-Live: Voice Model for Lifelike AI Conversations https://24-ai.news/en/news/2026-07-08/openai-swe-bench-pro-critique/ 24 AI en 2026-07-08T00:00:00Z OpenAI Challenges SWE-Bench Pro: The Leading AI Coding Benchmark Has Reliability Problems https://24-ai.news/en/news/2026-07-08/pytorch-2-13-release/ 24 AI en 2026-07-08T00:00:00Z PyTorch 2.13: Up to 4× Less GPU Memory for LLM Training and 12.3× Faster FlexAttention https://24-ai.news/de/news/2026-07-08/anthropic-claude-code-2-1-205/ 24 AI de 2026-07-08T00:00:00Z Claude Code v2.1.205: Sitzungssicherheit, Fehlerbehebungen und 400 MB RAM-Einsparung https://24-ai.news/de/news/2026-07-08/anthropic-gram-dual-use-offswitch/ 24 AI de 2026-07-08T00:00:00Z Anthropic entwickelt „Off Switch” für gefährliches Wissen: GRAM isoliert Dual-Use-Fähigkeiten in entfernbare Module https://24-ai.news/de/news/2026-07-08/arxiv-doomed-early-abort/ 24 AI de 2026-07-08T00:00:00Z Probe-Kaskade sagt Misserfolg des KI-Agenten aus verborgenen Aktivierungen im ersten Schritt voraus und spart bis zu 47 % der Token https://24-ai.news/de/news/2026-07-08/arxiv-dt-guard-safety/ 24 AI de 2026-07-08T00:00:00Z DT-Guard: 4B-Parameter-Modell übertrifft 8B-Guardrails durch Trennung von Reasoning-Supervision und Inferenzgeschwindigkeit https://24-ai.news/de/news/2026-07-08/arxiv-incontext-search-theory/ 24 AI de 2026-07-08T00:00:00Z Formale Theorie definiert präzise, wann iterative LLM-Reflexion exponentielle Verbesserungen bringt https://24-ai.news/de/news/2026-07-08/aws-claude-apps-gateway/ 24 AI de 2026-07-08T00:00:00Z AWS führt Claude Apps Gateway ein: Zentrales Management von Claude-Tools für Enterprise-Teams https://24-ai.news/de/news/2026-07-08/cncf-cloud-native-ai-storage/ 24 AI de 2026-07-08T00:00:00Z CNCF White Paper: Datenspeicherung bleibt größtes Hindernis für Cloud-Native-KI im Maßstab https://24-ai.news/de/news/2026-07-08/ibm-redhat-lightwell/ 24 AI de 2026-07-08T00:00:00Z IBM und Red Hat erweitern Lightwell – 6.500+ bereinigte Open-Source-Abhängigkeiten und Clearinghouse für Finanzen https://24-ai.news/de/news/2026-07-08/langchain-nvidia-nemoclaw/ 24 AI de 2026-07-08T00:00:00Z LangChain und NVIDIA launchen NemoClaw: Offener Agenten-Stack erzielt 10× niedrigere Kosten als Konkurrenz https://24-ai.news/de/news/2026-07-08/microsoft-flint-viz-language/ 24 AI de 2026-07-08T00:00:00Z Flint — Microsofts Open-Source-Sprache, die Visualisierungen für KI-Agenten kompiliert https://24-ai.news/de/news/2026-07-08/mistral-robostral-navigate/ 24 AI de 2026-07-08T00:00:00Z Mistral Robostral Navigate: Roboter-KI navigiert nur mit RGB-Kamera https://24-ai.news/de/news/2026-07-08/openai-government-national-security/ 24 AI de 2026-07-08T00:00:00Z OpenAI veröffentlicht Grundsätze für Partnerschaften mit Regierungen und Sicherheitsbehörden https://24-ai.news/de/news/2026-07-08/openai-gpt-live/ 24 AI de 2026-07-08T00:00:00Z OpenAI lanciert GPT-Live: Sprachmodell für lebensechte KI-Konversationen https://24-ai.news/de/news/2026-07-08/openai-swe-bench-pro-critique/ 24 AI de 2026-07-08T00:00:00Z OpenAI hinterfragt SWE-Bench Pro: Führender KI-Coding-Benchmark hat Zuverlässigkeitsprobleme https://24-ai.news/de/news/2026-07-08/pytorch-2-13-release/ 24 AI de 2026-07-08T00:00:00Z PyTorch 2.13: bis zu 4× weniger GPU-Speicher beim LLM-Training und 12,3× schnelleres FlexAttention https://24-ai.news/zh/news/2026-07-08/anthropic-claude-code-2-1-205/ 24 AI zh 2026-07-08T00:00:00Z Claude Code v2.1.205：会话安全、错误修复与节省400 MB内存 https://24-ai.news/zh/news/2026-07-08/anthropic-gram-dual-use-offswitch/ 24 AI zh 2026-07-08T00:00:00Z Anthropic开发危险知识「关闭开关」：GRAM将两用能力隔离为可移除模块 https://24-ai.news/zh/news/2026-07-08/arxiv-doomed-early-abort/ 24 AI zh 2026-07-08T00:00:00Z 探针级联从第一步隐藏激活预测AI智能体失败，节省高达47%的token https://24-ai.news/zh/news/2026-07-08/arxiv-dt-guard-safety/ 24 AI zh 2026-07-08T00:00:00Z DT-Guard：40亿参数模型通过分离推理监督与推理速度，超越80亿参数安全护栏 https://24-ai.news/zh/news/2026-07-08/arxiv-incontext-search-theory/ 24 AI zh 2026-07-08T00:00:00Z 形式理论精确界定：LLM迭代反思何时带来指数级性能提升？ https://24-ai.news/zh/news/2026-07-08/aws-claude-apps-gateway/ 24 AI zh 2026-07-08T00:00:00Z AWS推出Claude Apps Gateway：企业团队的Claude工具集中管理平台 https://24-ai.news/zh/news/2026-07-08/cncf-cloud-native-ai-storage/ 24 AI zh 2026-07-08T00:00:00Z CNCF白皮书：数据存储仍是云原生AI大规模落地的主要瓶颈 https://24-ai.news/zh/news/2026-07-08/ibm-redhat-lightwell/ 24 AI zh 2026-07-08T00:00:00Z IBM与Red Hat扩展Lightwell——6500+修复开源依赖项及金融行业信息交换中心 https://24-ai.news/zh/news/2026-07-08/langchain-nvidia-nemoclaw/ 24 AI zh 2026-07-08T00:00:00Z LangChain与NVIDIA发布NemoClaw：开放智能体栈成本比竞争对手低10倍 https://24-ai.news/zh/news/2026-07-08/microsoft-flint-viz-language/ 24 AI zh 2026-07-08T00:00:00Z Flint——微软开源的AI智能体图表编译语言 https://24-ai.news/zh/news/2026-07-08/mistral-robostral-navigate/ 24 AI zh 2026-07-08T00:00:00Z Mistral Robostral Navigate：仅用RGB摄像头导航的机器人AI模型 https://24-ai.news/zh/news/2026-07-08/openai-government-national-security/ 24 AI zh 2026-07-08T00:00:00Z OpenAI发布与政府及安全机构合作的原则声明 https://24-ai.news/zh/news/2026-07-08/openai-gpt-live/ 24 AI zh 2026-07-08T00:00:00Z OpenAI发布GPT-Live：面向生动自然对话的语音AI模型 https://24-ai.news/zh/news/2026-07-08/openai-swe-bench-pro-critique/ 24 AI zh 2026-07-08T00:00:00Z OpenAI质疑SWE-Bench Pro：AI编程领域主流基准存在可靠性问题 https://24-ai.news/zh/news/2026-07-08/pytorch-2-13-release/ 24 AI zh 2026-07-08T00:00:00Z PyTorch 2.13：LLM训练GPU显存最高降低4倍，FlexAttention提速12.3倍 https://24-ai.news/ja/news/2026-07-08/anthropic-claude-code-2-1-205/ 24 AI ja 2026-07-08T00:00:00Z Claude Code v2.1.205：セッションセキュリティ、バグ修正、RAM使用量400MB削減 https://24-ai.news/ja/news/2026-07-08/anthropic-gram-dual-use-offswitch/ 24 AI ja 2026-07-08T00:00:00Z Anthropic、危険な知識の「オフスイッチ」を開発：GRAMが二重利用可能な能力を除去可能なモジュールに分離 https://24-ai.news/ja/news/2026-07-08/arxiv-doomed-early-abort/ 24 AI ja 2026-07-08T00:00:00Z プローブのカスケードが第1ステップの隠れた活性化からAIエージェントの失敗を予測し、トークンを最大47%節約 https://24-ai.news/ja/news/2026-07-08/arxiv-dt-guard-safety/ 24 AI ja 2026-07-08T00:00:00Z DT-Guard：4Bパラメータモデルが推論スーパービジョンと推論速度の分離で8Bガードレールを凌駕 https://24-ai.news/ja/news/2026-07-08/arxiv-incontext-search-theory/ 24 AI ja 2026-07-08T00:00:00Z 反復的な反省がLLMに指数関数的な改善をもたらす条件を形式理論が正確に定義 https://24-ai.news/ja/news/2026-07-08/aws-claude-apps-gateway/ 24 AI ja 2026-07-08T00:00:00Z AWSがClaude Apps Gatewayを発表：エンタープライズチーム向けClaudeツールの一元管理 https://24-ai.news/ja/news/2026-07-08/cncf-cloud-native-ai-storage/ 24 AI ja 2026-07-08T00:00:00Z CNCFホワイトペーパー：データストレージがスケールするクラウドネイティブAIの主要ボトルネックに https://24-ai.news/ja/news/2026-07-08/ibm-redhat-lightwell/ 24 AI ja 2026-07-08T00:00:00Z IBMとRed HatがLightwellを拡充——6,500以上の修正済みオープンソース依存関係と金融向けクリアリングハウス https://24-ai.news/ja/news/2026-07-08/langchain-nvidia-nemoclaw/ 24 AI ja 2026-07-08T00:00:00Z LangChainとNVIDIAがNemoClawを発表：オープンエージェントスタックが競合の10倍低コストを実現 https://24-ai.news/ja/news/2026-07-08/microsoft-flint-viz-language/ 24 AI ja 2026-07-08T00:00:00Z Flint——AIエージェント向けビジュアライゼーションをコンパイルするMicrosoftのオープンソース言語 https://24-ai.news/ja/news/2026-07-08/mistral-robostral-navigate/ 24 AI ja 2026-07-08T00:00:00Z Mistral Robostral Navigate：RGBカメラ1台だけで自律ナビゲーションを実現するロボットAI https://24-ai.news/ja/news/2026-07-08/openai-government-national-security/ 24 AI ja 2026-07-08T00:00:00Z OpenAI、政府・安全保障機関との連携に関する原則を発表 https://24-ai.news/ja/news/2026-07-08/openai-gpt-live/ 24 AI ja 2026-07-08T00:00:00Z OpenAI、会話AIの新音声モデル「GPT-Live」をChatGPT Voiceに初日から統合して発表 https://24-ai.news/ja/news/2026-07-08/openai-swe-bench-pro-critique/ 24 AI ja 2026-07-08T00:00:00Z OpenAI、SWE-Bench Proの信頼性に疑問：AI coding評価の主要ベンチマークに問題あり https://24-ai.news/ja/news/2026-07-08/pytorch-2-13-release/ 24 AI ja 2026-07-08T00:00:00Z PyTorch 2.13：LLMトレーニングのGPUメモリを最大4分の1に削減、FlexAttentionが12.3倍高速化 https://24-ai.news/ko/news/2026-07-08/anthropic-claude-code-2-1-205/ 24 AI ko 2026-07-08T00:00:00Z Claude Code v2.1.205: 세션 보안, 버그 수정, 400MB RAM 절약 https://24-ai.news/ko/news/2026-07-08/anthropic-gram-dual-use-offswitch/ 24 AI ko 2026-07-08T00:00:00Z Anthropic, 위험 지식의 「오프 스위치」 개발: GRAM, 이중 사용 능력을 제거 가능한 모듈에 격리 https://24-ai.news/ko/news/2026-07-08/arxiv-doomed-early-abort/ 24 AI ko 2026-07-08T00:00:00Z 프로브 캐스케이드, 첫 단계에서 AI 에이전트 실패를 은닉 활성화로 예측하고 토큰 최대 47% 절약 https://24-ai.news/ko/news/2026-07-08/arxiv-dt-guard-safety/ 24 AI ko 2026-07-08T00:00:00Z DT-Guard: 추론 감독과 추론 속도를 분리해 8B 가드레일을 능가하는 4B 파라미터 모델 https://24-ai.news/ko/news/2026-07-08/arxiv-incontext-search-theory/ 24 AI ko 2026-07-08T00:00:00Z 공식 이론, LLM의 반복적 반성이 기하급수적 개선을 가져오는 시점을 정확히 정의하다 https://24-ai.news/ko/news/2026-07-08/aws-claude-apps-gateway/ 24 AI ko 2026-07-08T00:00:00Z AWS, Claude Apps Gateway 도입: 엔터프라이즈 팀을 위한 Claude 도구의 중앙 관리 https://24-ai.news/ko/news/2026-07-08/cncf-cloud-native-ai-storage/ 24 AI ko 2026-07-08T00:00:00Z CNCF 백서: 데이터 스토리지가 대규모 클라우드 네이티브 AI의 주요 장벽으로 남다 https://24-ai.news/ko/news/2026-07-08/ibm-redhat-lightwell/ 24 AI ko 2026-07-08T00:00:00Z IBM과 Red Hat, Lightwell 확장 — 6,500개 이상의 검증된 오픈소스 의존성과 금융 부문 클리어링하우스 https://24-ai.news/ko/news/2026-07-08/langchain-nvidia-nemoclaw/ 24 AI ko 2026-07-08T00:00:00Z LangChain과 NVIDIA, NemoClaw 출시: 오픈 에이전트 스택, 경쟁사 대비 10배 낮은 비용 달성 https://24-ai.news/ko/news/2026-07-08/microsoft-flint-viz-language/ 24 AI ko 2026-07-08T00:00:00Z Flint — AI 에이전트를 위한 시각화를 컴파일하는 Microsoft의 오픈소스 언어 https://24-ai.news/ko/news/2026-07-08/mistral-robostral-navigate/ 24 AI ko 2026-07-08T00:00:00Z Mistral Robostral Navigate: RGB 카메라만으로 탐색하는 로봇 AI https://24-ai.news/ko/news/2026-07-08/openai-government-national-security/ 24 AI ko 2026-07-08T00:00:00Z OpenAI, 정부 및 안보 기관과의 협력 원칙을 공표하다 https://24-ai.news/ko/news/2026-07-08/openai-gpt-live/ 24 AI ko 2026-07-08T00:00:00Z OpenAI, 실시간 AI 대화를 위한 음성 모델 GPT-Live 출시 https://24-ai.news/ko/news/2026-07-08/openai-swe-bench-pro-critique/ 24 AI ko 2026-07-08T00:00:00Z OpenAI, SWE-Bench Pro를 재검토하다: AI 코딩 선두 벤치마크의 신뢰성 문제 https://24-ai.news/ko/news/2026-07-08/pytorch-2-13-release/ 24 AI ko 2026-07-08T00:00:00Z PyTorch 2.13: LLM 학습 GPU 메모리 최대 4배 절감, FlexAttention 12.3배 빨라져 https://24-ai.news/hr/news/2026-07-07/aisi-cloud-misconfigurations-frontier-ai/ 24 AI hr 2026-07-07T00:00:00Z AISI: Frontier AI modeli otkrili kritične cloud ranjivosti za manje od £150 https://24-ai.news/hr/news/2026-07-07/amd-occupancy-math-mi355x/ 24 AI hr 2026-07-07T00:00:00Z AMD objašnjava occupancy matematiku za MI355X: visoka popunjenost nije uvjet za vršni throughput https://24-ai.news/hr/news/2026-07-07/anthropic-claude-code-2-1-203/ 24 AI hr 2026-07-07T00:00:00Z Claude Code v2.1.203 — upozorenje o isteku sesije, manji binarij, brži start https://24-ai.news/hr/news/2026-07-07/arxiv-agent-data-injection-attacks/ 24 AI hr 2026-07-07T00:00:00Z Agent Data Injection: Nova klasa napada koja zaobilazi obrane AI agenata https://24-ai.news/hr/news/2026-07-07/arxiv-agentgym2-benchmark/ 24 AI hr 2026-07-07T00:00:00Z AgentGym2: Novi benchmark testira LLM agente u de-idealiziranim uvjetima https://24-ai.news/hr/news/2026-07-07/arxiv-llm-as-verifier/ 24 AI hr 2026-07-07T00:00:00Z LLM-as-a-Verifier: Verifikacija kao nova, četvrta os skaliranja jezičnih modela https://24-ai.news/hr/news/2026-07-07/arxiv-weak-to-strong-direct-opd/ 24 AI hr 2026-07-07T00:00:00Z Direct-OPD: Kako slabiji model može trenirati jačeg bez skupog RL-a https://24-ai.news/hr/news/2026-07-07/eu-action-plan-cybersecurity-ai/ 24 AI hr 2026-07-07T00:00:00Z EU Akcijski plan za kibernetičku sigurnost i AI: Komisija koordinira obranu bez novih zakona https://24-ai.news/hr/news/2026-07-07/github-copilot-app-general-availability/ 24 AI hr 2026-07-07T00:00:00Z GitHub Copilot App dostupan svima — od besplatnog plana do Enterprise https://24-ai.news/hr/news/2026-07-07/google-managed-agents-gemini-api/ 24 AI hr 2026-07-07T00:00:00Z Google proširuje Managed Agents u Gemini API-ju za produkcijska okruženja https://24-ai.news/hr/news/2026-07-07/google-traffic-congestion-routing/ 24 AI hr 2026-07-07T00:00:00Z Google Research: Preusmjeravanje manje od 2% vožnji smanjuje gradske gužve i emisije CO2 https://24-ai.news/hr/news/2026-07-07/langchain-agents-data-mining/ 24 AI hr 2026-07-07T00:00:00Z LangChain: poboljšavanje agenata je problem rudarenja podataka https://24-ai.news/hr/news/2026-07-07/langchain-schneider-llmops-langsmith/ 24 AI hr 2026-07-07T00:00:00Z Kako je Schneider Electric izgradio LLMOps temelje za 60+ AI agenata s LangSmithom https://24-ai.news/hr/news/2026-07-07/meta-muse-image-video/ 24 AI hr 2026-07-07T00:00:00Z Meta lansira Muse Image i Muse Video: agentski AI koji sam ispravlja vlastite greške https://24-ai.news/hr/news/2026-07-07/nvidia-vera-cpu-single-threaded/ 24 AI hr 2026-07-07T00:00:00Z NVIDIA Vera: zašto agentski AI treba brže jezgre, a ne više jezgri https://24-ai.news/en/news/2026-07-07/aisi-cloud-misconfigurations-frontier-ai/ 24 AI en 2026-07-07T00:00:00Z AISI: Frontier AI Models Found Critical Cloud Vulnerabilities for Under £150 https://24-ai.news/en/news/2026-07-07/amd-occupancy-math-mi355x/ 24 AI en 2026-07-07T00:00:00Z AMD Explains Occupancy Math for the MI355X: High Occupancy Is Not Required for Peak Throughput https://24-ai.news/en/news/2026-07-07/anthropic-claude-code-2-1-203/ 24 AI en 2026-07-07T00:00:00Z Claude Code v2.1.203 — Session Expiry Warning, Smaller Binary, Faster Startup https://24-ai.news/en/news/2026-07-07/arxiv-agent-data-injection-attacks/ 24 AI en 2026-07-07T00:00:00Z Agent Data Injection: A New Attack Class That Bypasses AI Agent Defenses https://24-ai.news/en/news/2026-07-07/arxiv-agentgym2-benchmark/ 24 AI en 2026-07-07T00:00:00Z AgentGym2: New Benchmark Tests LLM Agents Under De-Idealized Conditions https://24-ai.news/en/news/2026-07-07/arxiv-llm-as-verifier/ 24 AI en 2026-07-07T00:00:00Z LLM-as-a-Verifier: Verification as a New Fourth Scaling Axis for Language Models https://24-ai.news/en/news/2026-07-07/arxiv-weak-to-strong-direct-opd/ 24 AI en 2026-07-07T00:00:00Z Direct-OPD: How a Weaker Model Can Train a Stronger One Without Expensive RL https://24-ai.news/en/news/2026-07-07/eu-action-plan-cybersecurity-ai/ 24 AI en 2026-07-07T00:00:00Z EU Action Plan on Cybersecurity and AI: Commission Coordinates Defense Without New Laws https://24-ai.news/en/news/2026-07-07/github-copilot-app-general-availability/ 24 AI en 2026-07-07T00:00:00Z GitHub Copilot App Available to Everyone — From Free Plan to Enterprise https://24-ai.news/en/news/2026-07-07/google-managed-agents-gemini-api/ 24 AI en 2026-07-07T00:00:00Z Google Expands Managed Agents in the Gemini API for Production Environments https://24-ai.news/en/news/2026-07-07/google-traffic-congestion-routing/ 24 AI en 2026-07-07T00:00:00Z Google Research: Rerouting Less Than 2% of Trips Reduces Urban Congestion and CO2 Emissions https://24-ai.news/en/news/2026-07-07/langchain-agents-data-mining/ 24 AI en 2026-07-07T00:00:00Z LangChain: Improving Agents Is a Data Mining Problem https://24-ai.news/en/news/2026-07-07/langchain-schneider-llmops-langsmith/ 24 AI en 2026-07-07T00:00:00Z How Schneider Electric Built LLMOps Foundations for 60+ AI Agents with LangSmith https://24-ai.news/en/news/2026-07-07/meta-muse-image-video/ 24 AI en 2026-07-07T00:00:00Z Meta Launches Muse Image and Muse Video: Agentic AI That Self-Corrects Its Own Mistakes https://24-ai.news/en/news/2026-07-07/nvidia-vera-cpu-single-threaded/ 24 AI en 2026-07-07T00:00:00Z NVIDIA Vera: Why Agentic AI Needs Faster Cores, Not More Cores https://24-ai.news/de/news/2026-07-07/aisi-cloud-misconfigurations-frontier-ai/ 24 AI de 2026-07-07T00:00:00Z AISI: Frontier-KI-Modelle entdeckten kritische Cloud-Schwachstellen für unter £150 https://24-ai.news/de/news/2026-07-07/amd-occupancy-math-mi355x/ 24 AI de 2026-07-07T00:00:00Z AMD erklärt Occupancy-Mathematik für MI355X: Hohe Auslastung ist keine Voraussetzung für Peak-Throughput https://24-ai.news/de/news/2026-07-07/anthropic-claude-code-2-1-203/ 24 AI de 2026-07-07T00:00:00Z Claude Code v2.1.203 — Session-Ablaufwarnung, kleineres Binary, schnellerer Start https://24-ai.news/de/news/2026-07-07/arxiv-agent-data-injection-attacks/ 24 AI de 2026-07-07T00:00:00Z Agent Data Injection: Neue Angriffskategorie umgeht Schutzmaßnahmen für KI-Agenten https://24-ai.news/de/news/2026-07-07/arxiv-agentgym2-benchmark/ 24 AI de 2026-07-07T00:00:00Z AgentGym2: Neuer Benchmark testet LLM-Agenten unter de-idealisierten Bedingungen https://24-ai.news/de/news/2026-07-07/arxiv-llm-as-verifier/ 24 AI de 2026-07-07T00:00:00Z LLM-as-a-Verifier: Verifikation als neue, vierte Skalierungsachse für Sprachmodelle https://24-ai.news/de/news/2026-07-07/arxiv-weak-to-strong-direct-opd/ 24 AI de 2026-07-07T00:00:00Z Direct-OPD: Wie ein schwächeres Modell ein stärkeres ohne teures RL trainieren kann https://24-ai.news/de/news/2026-07-07/eu-action-plan-cybersecurity-ai/ 24 AI de 2026-07-07T00:00:00Z EU-Aktionsplan für Cybersicherheit und KI: Kommission koordiniert Abwehr ohne neue Gesetze https://24-ai.news/de/news/2026-07-07/github-copilot-app-general-availability/ 24 AI de 2026-07-07T00:00:00Z GitHub Copilot App für alle verfügbar — vom kostenlosen Plan bis Enterprise https://24-ai.news/de/news/2026-07-07/google-managed-agents-gemini-api/ 24 AI de 2026-07-07T00:00:00Z Google erweitert Managed Agents in der Gemini API für Produktionsumgebungen https://24-ai.news/de/news/2026-07-07/google-traffic-congestion-routing/ 24 AI de 2026-07-07T00:00:00Z Google Research: Weniger als 2 % umgeleitete Fahrten verringern Stadtverkehr und CO2-Emissionen https://24-ai.news/de/news/2026-07-07/langchain-agents-data-mining/ 24 AI de 2026-07-07T00:00:00Z LangChain: Agenten verbessern ist ein Data-Mining-Problem https://24-ai.news/de/news/2026-07-07/langchain-schneider-llmops-langsmith/ 24 AI de 2026-07-07T00:00:00Z Wie Schneider Electric LLMOps-Grundlagen für 60+ KI-Agenten mit LangSmith aufgebaut hat https://24-ai.news/de/news/2026-07-07/meta-muse-image-video/ 24 AI de 2026-07-07T00:00:00Z Meta launcht Muse Image und Muse Video: Agentische KI, die eigene Fehler selbst korrigiert https://24-ai.news/de/news/2026-07-07/nvidia-vera-cpu-single-threaded/ 24 AI de 2026-07-07T00:00:00Z NVIDIA Vera: Warum agentische KI schnellere Kerne braucht, nicht mehr Kerne https://24-ai.news/zh/news/2026-07-07/aisi-cloud-misconfigurations-frontier-ai/ 24 AI zh 2026-07-07T00:00:00Z AISI：前沿AI模型以不足150英镑的成本发现关键云端漏洞 https://24-ai.news/zh/news/2026-07-07/amd-occupancy-math-mi355x/ 24 AI zh 2026-07-07T00:00:00Z AMD详解MI355X的占用率数学：高占用率并非峰值吞吐量的必要条件 https://24-ai.news/zh/news/2026-07-07/anthropic-claude-code-2-1-203/ 24 AI zh 2026-07-07T00:00:00Z Claude Code v2.1.203 — 会话过期警告、更小的二进制文件、更快的启动速度 https://24-ai.news/zh/news/2026-07-07/arxiv-agent-data-injection-attacks/ 24 AI zh 2026-07-07T00:00:00Z 智能体数据注入攻击：绕过AI智能体防御的全新攻击类别 https://24-ai.news/zh/news/2026-07-07/arxiv-agentgym2-benchmark/ 24 AI zh 2026-07-07T00:00:00Z AgentGym2：全新基准在去理想化条件下测试LLM智能体 https://24-ai.news/zh/news/2026-07-07/arxiv-llm-as-verifier/ 24 AI zh 2026-07-07T00:00:00Z LLM-as-a-Verifier：验证作为语言模型扩展的全新第四维度 https://24-ai.news/zh/news/2026-07-07/arxiv-weak-to-strong-direct-opd/ 24 AI zh 2026-07-07T00:00:00Z Direct-OPD：较弱模型如何在无需昂贵强化学习的情况下训练更强模型 https://24-ai.news/zh/news/2026-07-07/eu-action-plan-cybersecurity-ai/ 24 AI zh 2026-07-07T00:00:00Z 欧盟网络安全与AI行动计划：欧盟委员会无需新立法即可协调防御 https://24-ai.news/zh/news/2026-07-07/github-copilot-app-general-availability/ 24 AI zh 2026-07-07T00:00:00Z GitHub Copilot App向所有人开放——从免费计划到Enterprise https://24-ai.news/zh/news/2026-07-07/google-managed-agents-gemini-api/ 24 AI zh 2026-07-07T00:00:00Z Google为Gemini API中的Managed Agents扩展生产级功能 https://24-ai.news/zh/news/2026-07-07/google-traffic-congestion-routing/ 24 AI zh 2026-07-07T00:00:00Z Google研究：引导不足2%的车辆可减少城市交通拥堵和CO2排放 https://24-ai.news/zh/news/2026-07-07/langchain-agents-data-mining/ 24 AI zh 2026-07-07T00:00:00Z LangChain：改进智能体是一个数据挖掘问题 https://24-ai.news/zh/news/2026-07-07/langchain-schneider-llmops-langsmith/ 24 AI zh 2026-07-07T00:00:00Z 施耐德电气如何借助LangSmith为60余个AI智能体构建企业级LLMOps基础 https://24-ai.news/zh/news/2026-07-07/meta-muse-image-video/ 24 AI zh 2026-07-07T00:00:00Z Meta发布Muse Image和Muse Video：能自主纠错的智能体AI https://24-ai.news/zh/news/2026-07-07/nvidia-vera-cpu-single-threaded/ 24 AI zh 2026-07-07T00:00:00Z NVIDIA Vera：为何智能体AI需要更快的核心，而非更多核心 https://24-ai.news/ja/news/2026-07-07/aisi-cloud-misconfigurations-frontier-ai/ 24 AI ja 2026-07-07T00:00:00Z AISI：フロンティアAIが£150未満でクリティカルなクラウド脆弱性を発見 https://24-ai.news/ja/news/2026-07-07/amd-occupancy-math-mi355x/ 24 AI ja 2026-07-07T00:00:00Z AMDがMI355XのGPU占有率の計算を解説：高い占有率がピークスループットの条件ではない https://24-ai.news/ja/news/2026-07-07/anthropic-claude-code-2-1-203/ 24 AI ja 2026-07-07T00:00:00Z Claude Code v2.1.203――セッション期限切れ警告、小さなバイナリ、高速起動 https://24-ai.news/ja/news/2026-07-07/arxiv-agent-data-injection-attacks/ 24 AI ja 2026-07-07T00:00:00Z Agent Data Injection：AIエージェントの防御を回避する新しい攻撃クラス https://24-ai.news/ja/news/2026-07-07/arxiv-agentgym2-benchmark/ 24 AI ja 2026-07-07T00:00:00Z AgentGym2：新ベンチマークがLLMエージェントを非理想的な条件でテスト https://24-ai.news/ja/news/2026-07-07/arxiv-llm-as-verifier/ 24 AI ja 2026-07-07T00:00:00Z LLM-as-a-Verifier：検証は言語モデルの新しい第4のスケーリング軸 https://24-ai.news/ja/news/2026-07-07/arxiv-weak-to-strong-direct-opd/ 24 AI ja 2026-07-07T00:00:00Z Direct-OPD：高コストのRLなしに弱いモデルが強いモデルを訓練できる方法 https://24-ai.news/ja/news/2026-07-07/eu-action-plan-cybersecurity-ai/ 24 AI ja 2026-07-07T00:00:00Z EUサイバーセキュリティ・AI行動計画：欧州委員会が新立法なしで防衛を調整 https://24-ai.news/ja/news/2026-07-07/github-copilot-app-general-availability/ 24 AI ja 2026-07-07T00:00:00Z GitHub Copilot Appが全ユーザーに提供開始――無料プランからEnterpriseまで https://24-ai.news/ja/news/2026-07-07/google-managed-agents-gemini-api/ 24 AI ja 2026-07-07T00:00:00Z GoogleがGemini APIのManaged Agentsを本番環境向けに拡張 https://24-ai.news/ja/news/2026-07-07/google-traffic-congestion-routing/ 24 AI ja 2026-07-07T00:00:00Z Google Research：走行の2%未満をルート変更するだけで都市の渋滞とCO2排出が減少 https://24-ai.news/ja/news/2026-07-07/langchain-agents-data-mining/ 24 AI ja 2026-07-07T00:00:00Z LangChain：エージェントの改善はデータマイニングの問題だ https://24-ai.news/ja/news/2026-07-07/langchain-schneider-llmops-langsmith/ 24 AI ja 2026-07-07T00:00:00Z Schneider ElectricがLangSmithで60以上のAIエージェントのLLMOpsの基盤を構築した方法 https://24-ai.news/ja/news/2026-07-07/meta-muse-image-video/ 24 AI ja 2026-07-07T00:00:00Z MetaがMuse ImageとMuse Videoを発表――自ら誤りを修正するエージェント型AI https://24-ai.news/ja/news/2026-07-07/nvidia-vera-cpu-single-threaded/ 24 AI ja 2026-07-07T00:00:00Z NVIDIA Vera：エージェント型AIに必要なのは多くのコアではなく速いコアだ https://24-ai.news/ko/news/2026-07-07/aisi-cloud-misconfigurations-frontier-ai/ 24 AI ko 2026-07-07T00:00:00Z AISI: 프론티어 AI 모델, £150 미만으로 심각한 클라우드 취약점 발견 https://24-ai.news/ko/news/2026-07-07/amd-occupancy-math-mi355x/ 24 AI ko 2026-07-07T00:00:00Z AMD, MI355X occupancy 수학 설명 — 높은 점유율이 최고 처리량의 조건이 아니다 https://24-ai.news/ko/news/2026-07-07/anthropic-claude-code-2-1-203/ 24 AI ko 2026-07-07T00:00:00Z Claude Code v2.1.203 — 세션 만료 경고, 더 작은 바이너리, 빠른 시작 https://24-ai.news/ko/news/2026-07-07/arxiv-agent-data-injection-attacks/ 24 AI ko 2026-07-07T00:00:00Z Agent Data Injection: AI 에이전트 방어를 우회하는 새로운 공격 클래스 https://24-ai.news/ko/news/2026-07-07/arxiv-agentgym2-benchmark/ 24 AI ko 2026-07-07T00:00:00Z AgentGym2: 새 벤치마크, 비이상화된 조건에서 LLM 에이전트를 테스트 https://24-ai.news/ko/news/2026-07-07/arxiv-llm-as-verifier/ 24 AI ko 2026-07-07T00:00:00Z LLM-as-a-Verifier: 검증, 언어 모델 스케일링의 새 네 번째 축 https://24-ai.news/ko/news/2026-07-07/arxiv-weak-to-strong-direct-opd/ 24 AI ko 2026-07-07T00:00:00Z Direct-OPD: 약한 모델이 비싼 RL 없이 강한 모델을 훈련하는 방법 https://24-ai.news/ko/news/2026-07-07/eu-action-plan-cybersecurity-ai/ 24 AI ko 2026-07-07T00:00:00Z EU 사이버보안·AI 행동계획: 유럽위원회, 새 법률 없이 방어 조율 https://24-ai.news/ko/news/2026-07-07/github-copilot-app-general-availability/ 24 AI ko 2026-07-07T00:00:00Z GitHub Copilot App, 모든 사용자에게 공개 — 무료 플랜부터 Enterprise까지 https://24-ai.news/ko/news/2026-07-07/google-managed-agents-gemini-api/ 24 AI ko 2026-07-07T00:00:00Z Google, 프로덕션 환경을 위한 Gemini API Managed Agents 확장 https://24-ai.news/ko/news/2026-07-07/google-traffic-congestion-routing/ 24 AI ko 2026-07-07T00:00:00Z Google Research: 전체 운행의 2% 미만을 우회시켜 도시 혼잡과 CO2 배출 감소 https://24-ai.news/ko/news/2026-07-07/langchain-agents-data-mining/ 24 AI ko 2026-07-07T00:00:00Z LangChain: 에이전트 개선은 데이터 마이닝 문제다 https://24-ai.news/ko/news/2026-07-07/langchain-schneider-llmops-langsmith/ 24 AI ko 2026-07-07T00:00:00Z Schneider Electric, LangSmith로 60개 이상의 AI 에이전트를 위한 LLMOps 기반 구축 https://24-ai.news/ko/news/2026-07-07/meta-muse-image-video/ 24 AI ko 2026-07-07T00:00:00Z Meta, Muse Image와 Muse Video 출시 — 자체 오류를 수정하는 에이전트 AI https://24-ai.news/ko/news/2026-07-07/nvidia-vera-cpu-single-threaded/ 24 AI ko 2026-07-07T00:00:00Z NVIDIA Vera: 에이전트 AI에 더 많은 코어가 아닌 더 빠른 코어가 필요한 이유