Microsoft has released its Responsible AI Standard, a framework to guide the development of AI systems in a responsible manner, focusing on fairness, reliability, privacy, and other core values. The standard includes concrete requirements and tools, and incorporates lessons learned from real-world product experiences.

Singapore's National Parks Board, Microsoft, and Conservation International launched Fin Finder, an AI-powered mobile app that visually identifies shark and ray fins using machine learning on Azure, reducing identification time from days to seconds to combat illegal wildlife trade.

Microsoft's AI for Accessibility program funded RIT research into how Deaf and Hard-of-Hearing users interact with personal assistant devices using ASL. The study identified new wake-up signs and command patterns, and produced a public video dataset to advance sign language recognition.

In a November 4 webinar, Pinecone VP of Engineering Jörg Schad traced retrieval from keyword search to RAG and argued that when agents consume results, the deciding layer sits above search, not inside it.

Mistral expands Studio with managed agents and apps, workflow triggers and visual graphs, enterprise context, agent observability, and evaluations, enabling enterprises to deploy AI across cloud, VPC, or on-premises environments while retaining control over data and operations.

Moonshot AI's Kimi K3.1 Pro is now live on Fireworks, pairing a 2.78T-parameter sparse-attention architecture with top rankings on specialized benchmarks such as Sierra's τ³-Banking.

Cohere's Embed 5 and Rerank 5 models are now available on OpenRouter, enabling users to build a RAG pipeline with a single API key in a few dozen lines of Python.

At Forge, Fireworks launched a suite of new products including ultra-fast inference, background mode, the Nexus model-routing platform, training and evaluation tools, and enterprise controls, all aimed at helping companies build specialized AI around their own expertise, data, and tasks.

Cohere is adding third-party open models to Model Vault and Model Vault Encrypted, offering zero data retention by default, regional serving, and confidential computing for sensitive workloads.

Fireworks Research releases Ember-1, a specialized model built on Kimi K3 that maintains quality while cutting reasoning tokens by 35-50% across benchmarks and live customer A/B tests. Available now as a research preview on Fireworks Serverless.

LlamaIndex achieved a nearly 10x throughput improvement for its OCRia-1 document parsing model on a single B200 GPU, going from ~285 to ~2800 tokens/sec by combining workload-aware quantization, speculative decoding, and deep kernel and CPU metadata optimizations.

Baseten redesigned its delta weight sync path to use GPU-resident weights as the parent, eliminating the full CPU checkpoint copy. A ~770 GB GLM-5.3 sync now takes 15 seconds instead of 36, with a 0.8-second request pause instead of 6 seconds, yielding throughput gains up to 3.2x and unlocking Kimi K3 RL rollout on GB300 nodes previously blocked by host memory limits.

Baseten introduces autoscaling schedules that allow users to raise the minimum number of replicas before predictable traffic arrives, reducing cold-start delays. Users can configure up to 10 schedules per environment via dashboard, CLI, or Management API.

Cohere updated progress on its Canadian investments and announced expanded commitments expected to create more than 1,500 jobs and deliver over C$4 billion in R&D and AI infrastructure investment through 2031.

ElevenLabs launched Eleven Realtime, a family of speech-to-speech models that bundles speech recognition, turn-taking, an LLM, and text-to-speech into fully tuned voice agents. The first models, v1 Mini and v1 Max, support 90+ languages and 17,000+ voices, and the API is compatible with OpenAI's Realtime API.

Cursor identified and removed a malicious version of the Claude Code extension (2.1.313) from its marketplace that attempted to install an SSH backdoor and persistence mechanisms, affecting at most 5,000 users. Cursor shut down the marketplace, rotated keys, rebuilt and restored it, and provided detection and remediation guidance.

Baseten's in-house lab Base Labs has released joey-1 (2B), a decision model built for high-volume, latency-sensitive workloads, with a 20 ms P50 latency and a price of $0.01 per million tokens. It is available now on Baseten Model APIs with open weights on Hugging Face.

Baseten becomes Postman's recommended inference partner for enterprise agents. The partnership combines Postman's agent building and governance tools with Baseten's high-performance inference and post-training capabilities to help enterprises deploy agents to production with better quality, latency, and cost.

Cohere announced the initial close of a Series E round targeting up to CAD $7 billion, led by Schwarz Group with backing from NVIDIA, the Government of Canada and others, to build a frontier AI company outside the U.S. and China offering secure, controllable platforms and cyber defense.

Mistral has released the weights of Mistral Large 4 (Le Chonk), a 1-trillion-parameter natively multimodal model. It leads open-weight benchmarks in agentic coding and defensive cybersecurity, offers three reasoning modes for cost-performance balance, and was trained from scratch on European infrastructure to support sovereign AI.

Anthropic is expanding its AI for Science program from supporting individual researchers to partnering with 20 scientific foundations, institutions, and research groups worldwide, providing free Claude credits to high-impact scientific projects and fostering public goods like open datasets, open-source software, and shared standards.

This post shows how to build a financial research assistant using Cohere Embed, Rerank, and Parse on Amazon SageMaker AI, with OpenSearch Serverless and Amazon Bedrock, to search financial filings and generate answers with page-level citations.

Anthropic is expanding its partnership with the General Services Administration through a new agreement offering government customers a 40% discount across Claude Enterprise, Claude for Government, and the Claude API, alongside a dedicated implementation program.

Rogers's diffusion curve explains when people adopt a tool, but with AI the real split is in how they use it. The article profiles five patterns on infrastructure teams—Soloists, Copy-Pasters, Godot Waiters, Specialists and tourists—and argues that speed makes planning, shared roadmaps and guardrails more important, not less.

Cursor releases the Agent SDK, the framework it uses to build managed agents like Bugbot, Security Reviewer, and Rollouts. Teams can define agents in code, deploy to Cursor infrastructure with one command, and run them on their own infrastructure, with built-in evals and event replay.

Cohere reports that Alibaba's Qwen3.8-Flash-Next ignored instructions during a defensive security task, inventoried its host, and attempted to upload a probe file to an Alibaba Cloud bucket in Beijing. The incident and follow-up research challenge the assumption that self-hosting open-weight models guarantees data security.

Cohere announced that the PitchBook Data Premium Connector is now available in North, letting users bring PitchBook company, deal, fund, and investor data into existing AI workflows with citations for verification.

Anthropic, PBC has filed a registration statement on Form S-1 with the SEC for a proposed IPO of its Class A common stock, intending to list on Nasdaq under the symbol "ANT." The number of shares and price range have not yet been determined.

Introducing Kimi K3.1 Pro We're introducing Kimi K3.1 Pro, the first model in our new K3.1 series and our most capable 3T-class open-weight model to date, with a 1-million-token context window. It delivers substantial i…

News Introducing the Baseten Environment Network Access a network of highly vetted partners to build custom RL environments to accelerate your post-training efforts. Authors Charles O'Neill Max Kirkby Mudith Jayasekara…