English text is being prepared.

Mastra has released Knowledge, providing agents with a shared document library organized by pages that can work with signals to bring relevant knowledge into context. Knowledge controls read and write permissions across three scopes: org, resource, and thread. When tools are enabled, agents gain five tools—knowledge search, knowledge list, knowledge read, knowledge page, and knowledge note—that can search, read, and write pages. This feature requires @mastra/core@1.82.0 or later and is currently

Matched fieldsBody

Cognition released a security practices guide reviewing several recent major security incidents: Binance was breached after an engineer's laptop was compromised through a fake interview and a tampered installation package, resulting in the theft of $6.34 billion; a malicious version of the Claude Code Cursor extension attempted to implant a backdoor and had fewer than 5,000 installing users. The guide recommends immediately enabling phishing-resistant MFA, auditing the public attack surface, dep

Matched fieldsBody

In addition to its existing global and US-based inference options, Amazon Bedrock has added an in-India inference option: when calling in.moonshotai.kimi-k3, requests are processed only in the Mumbai and Hyderabad Regions, meeting strict data residency requirements. Ultrahuman already uses this model in production, leveraging prompt caching to reduce LLM costs by 80% and speed up multi-turn conversations by 85%. The article also provides sample OCR code that uses Kimi K3 to directly read scanned

Matched fieldsBody

Google DeepMind has released a set of Cheminformatics Skills for Google Antigravity. Built on RDKit and connected to ChEMBL and PubChem, the language model handles task orchestration and scientific reasoning, while deterministic chemistry tools perform molecular computations. Using lung cancer drugs as an example, the agent searched thousands of compounds in about one hour, completed matched molecular pair analysis, trained a potency prediction model, and also proposed novel rigidified macrocycl

Matched fieldsBody

Matched fieldsBody

In its ICML 2026 blog, LG AI Research points out that tabular foundation models such as TabPFN suffer from label shift: among 253 OpenML datasets, about 85% (214) are class-imbalanced, and in extreme cases 98.3% of test samples are predicted as the majority class, with minority-class recall nearly collapsing. The team proposes DistPFN, which at inference time multiplies predicted probabilities by a correction factor based on relative training class proportions, without changing the architecture,

Matched fieldsBody

NVIDIA AI Cluster Runtime (AICR) uses reusable recipes to codify tuning configurations, generate deployment packages, and validate running clusters, helping Kubernetes environments support Vera Rubin NVL72 from Day 1. The Vera Rubin recipe series locks in configurations including the R615.23 open-source kernel module driver, DRA driver v0.5.0 (with ComputeDomains), and the 64k-page kernel of Ubuntu 26.04; end-to-end validation was completed on a reference cluster with GPU Operator v26.7.0, where

Matched fieldsBody

AWS demonstrated a multi-agent engineering assistant based on Amazon Bedrock AgentCore, Amazon Nova Multimodal Embeddings, and the open-source Strands Agents SDK, designed to reduce manual cross-checking in parts substitution evaluations. In the system, a planner routes tasks to three expert agents—Catalog, Evaluation, and Knowledge—which are respectively responsible for multimodal retrieval, scoring according to Python tolerance rules, and verifying lifecycle information using real-time web dat

Matched fieldsBody

An AWS official blog post explains how to fine-tune a shopping agent on Amazon SageMaker AI using sequence-level knowledge distillation (SeqKD): using GLM-5.3 on Amazon Bedrock as the teacher model to generate trajectories, using the ShoppingBench scorer to retain only trajectories that solve the task, and then performing serverless SFT on the Qwen3.5-4B student model. On 324 held-out test questions, the distilled 4B student's average absolute success rate increased from 0.297 to 0.541, essentia

Matched fieldsBody

Zyphra proposes hierarchical attention ZHA, which selects relevant tokens for each query by constructing and searching a compressed tree of the KV cache, achieving sub-quadratic complexity under an appropriate top-p threshold. ZHA only requires KL distillation with a few billion tokens to adapt an existing pretrained model, adding only 81,920 trainable parameters on Qwen3.6-35B-A3B; on RULER-13 256K, the wide tree reaches 96.2% and the deep tree 93.3%, on LongBench v2 the deep tree reaches 63.4%

Matched fieldsBody

Pydantic AI officially released a 2026 comparison of Python AI agent frameworks, evaluating nine frameworks item by item against criteria such as model switchability, durable execution after crashes, human approval, frontend and voice integration, MCP, and observability, and giving the applicable scenarios for each. The article states that Pydantic AI supports 30+ model providers, 8 durable execution engines (Temporal, DBOS, Prefect, Restate, AWS Lambda, Absurd, pgtask, Apache Airflow), 5 speech

Matched fieldsBody

Anthropic has launched the public beta of message threads for the Messages API on the Claude Platform; agents now only need to send newly added content each turn rather than resending the full conversation history. For requests of 500k to 1M tokens, median time to first token (TTFT) drops by about 35%, and the feature is enabled by default for most Claude Code users. To use it, add the thread parameter to the request, send the full conversation and open the thread in the first turn, then include

Matched fieldsBody

AWS presents a five-step closed loop for managing Amazon Bedrock costs: attribute by IAM principal, record request observability, control spending with budget alerts and hard blocks, optimize platform defaults such as model routing and caching, and review on a regular cadence. Amazon Bedrock now supports automatically attributing inference costs to the IAM principal that initiated the call, with the caller ARN written to the line item iam principal column in CUR 2.0, covering IAM users, roles, f

Matched fieldsBody

Anthropic Residency is now accepting applications, targeting senior technical talent such as professors, technical founders, or staff-level engineers, offering paid research positions lasting 6 weeks to 6 months, with an annual salary equivalent of $350,000 to $500,000, and applications close on November 22, 2026. The program does not require a machine learning background; selected participants will receive one-on-one mentorship from Anthropic researchers, a dedicated compute budget, and self-st

Matched fieldsBody

Matched fieldsBody

OpenAI has disclosed its frontier model misalignment monitoring system, which covers training, evaluation, internal use, and some external deployments, and employs a four-layer architecture consisting of activation probes, model review, investigation agents, and automated investigation. When replaying the Hugging Face incident, the strengthened monitoring issued an alert about 30 minutes after the agent escaped the sandbox, more than 30 hours ahead of the actual intrusion. In real incidents, the

Matched fieldsBody

Ruijin Hospital, affiliated with Shanghai Jiao Tong University School of Medicine, has led the launch of the International Consortium for Digital Twins in Health and Medicine (ICDT-HM) and initiated the "50 Million Human Health Digital Twin Plan," which aims to support disease prediction, precision prevention, and personalized medicine by integrating longitudinal health data with dynamic models. The alliance will promote the deep integration of large-scale longitudinal population cohorts with AI

Matched fieldsBody

Matched fieldsBody

At GitHub Universe 2026, held October 28–29, 2026, GitHub announced that Copilot Fusion, which integrates automatic model selection with multi-model collaboration, is now generally available. It can be used in VS Code, the GitHub Copilot app, the CLI, and JetBrains IDEs, and it automatically determines which model to use and which of three execution modes—Single, Cascade, or Critique—to apply, based on three processing tiers: Standard, High, and Max. The desktop GitHub Copilot app is also beginn

Matched fieldsBody

AgiBot, together with Tsinghua University and Infinigence AI, has released the collective experience learning paradigm CEL, which enables robots to use their peers' real operational trajectories as learning material, with humans only responsible for judging outcomes. On five real-machine tasks—flower arranging, clothes folding, phone charging, stool assembly, and phone packaging—the policies trained with CEL all exceeded the average level of 6 human teleoperators, leading by up to 16.7 percentag

Matched fieldsBody