Browse independent reporting and analysis across AI models, research, technology, robotics, security, policy, ethics and the business of artificial intelligence.
Qwen-Drive 1.0 adapts Alibaba’s 4-billion-parameter multimodal model for 3D perception, driving questions and trajectory planning while keeping the base architecture intact.
Anthropic says Claude completed a 13-million-line Lean formalization of Fermat’s Last Theorem in 11 days, turning a celebrated mathematical proof into a machine-checkable artifact and raising new questions about scientific verification.
Google DeepMind’s first AI for the Planet accelerator in Asia-Pacific is supporting 16 teams working on biodiversity, agriculture and carbon markets, a test of whether frontier models can become dependable environmental infrastructure.
Mistral has raised €3 billion at a valuation above €21 billion, giving Europe’s leading independent AI lab more capital to train models, build infrastructure and test whether technological sovereignty can become a durable business.
AWS testing on Amazon EKS found that two configuration changes to Run:ai Model Streamer reduced repeat model startup time by 80% to 93%. The result targets an overlooked inference bottleneck: moving large weights into GPU memory before a pod can serve traffic.
AWS argues that autonomous agents require continuous behavioral monitoring, scoped identities and tiered automated response. Its framework extends existing cloud security practices to software that can authenticate, use tools and change tactics without waiting for a person.
A new user-managed login flow lets Bedrock Managed Knowledge Base connect to common workplace repositories with an employee's existing credentials. AWS says the option lowers prototype friction while preserving service-account setup for production.
Bedrock Managed Knowledge Base can now ingest ServiceNow knowledge articles and catalog items without a custom pipeline. Scope controls let teams limit crawls to selected knowledge bases, categories and service catalogs.
Anthropic's Claude Fable 5.1 is now available through Bedrock and Claude Platform on AWS for long-running coding, research and knowledge work. Its Covered Model status adds retention, safety-review and access requirements to the enterprise deployment decision.
AWS Agent Registry has launched under a dedicated namespace with approval and discovery controls for agents, MCP servers and skills. Preview users have until September 17 to move away from the earlier AgentCore namespace.
Two AWS reference implementations use AgentCore, Kiro and coding agents to generate database diagrams and review SQL changes for security issues. The examples put human checkpoints around construction work that models can now perform continuously.
AWS says Amazon Quick Automate can coordinate agents across departments, user interfaces and APIs, but its production guidance emphasizes narrow responsibilities, observable handoffs and human exception paths over open-ended autonomy.
AWS has published a LiteLLM gateway design that routes Codex through ECS and Bedrock with scoped identities, budgets, rate limits and telemetry. The architecture gives enterprises a control point without replacing the developer's coding interface.
AWS has documented a pattern-first migration of 20 healthcare and life-sciences agents to Bedrock AgentCore in about two and a half hours. The case study shows how parallel coding agents can compress platform migrations when a tested template comes first.
OpenAI chief scientist Jakub Pachocki argues that reasoning models should be treated as alien rather than human minds. His essay pairs confidence in machine intelligence with a warning that familiar language can conceal unfamiliar internal strategies.
OpenAI says it has reached its goal of building an automated research intern and is aiming next for a supervised automated researcher. The milestone raises practical questions about evaluation, oversight and how quickly AI can improve the systems used to build AI.
AWS has demonstrated a multimodal ordering assistant that serves text, voice notes and live calls through one WhatsApp business number. Shared memory lets the system recognize a customer across channels without merging the channel and ordering layers.
Intuit says its EWOK Agent can translate a plain-language request into a governed production failover using Amazon Bedrock. The system keeps policy checks, approvals and an audit trail around actions that can affect critical services.
AWS has published a pipeline for preprocessing dense PDFs and images with Textract before they enter a Bedrock knowledge base. The approach targets documents where tables, forms and page structure routinely break ordinary retrieval systems.
HyperPod InstantStart gives operators a shared control plane for cluster setup, training, inference and storage through a web interface or an AI agent. The release shows cloud infrastructure moving from assisted administration toward governed machine execution.
AWS and Nvidia have outlined a continuous training pipeline for robots and autonomous vehicles using Cosmos 3 and SageMaker HyperPod. The architecture treats synthetic data, post-training and closed-loop simulation as one recurring production system.
AWS has published a production design for scoring, consolidating and deleting stale AgentCore memories. The guidance turns forgetting into an operational control for agents that accumulate months of customer and workflow context.
OpenAI has released GPT-6 Astra to a limited group of enterprises before a wider paid-plan and API rollout. The new flagship can run longer professional workflows, but its critical cybersecurity capability has turned access controls and continuous monitoring into part of the product.
The Model Context Protocol's July specification removes mandatory sessions, sticky routing and the initialize handshake. AWS says the change makes serverless deployment a natural fit, though teams must preserve a compatibility lane for older clients.