Hugging Face and partners announced Project ProtoStar, an initiative to make suitable scientific datasets, models and code available in AI-ready formats on open infrastructure. Its founding partners include five universities, the Astera Institute and the Open Research Community Accelerator. The initiative also plans open-science challenges to demonstrate reuse …
At GitHub Universe 2026, GitHub announced a Copilot app for agent-oriented work across code and non-code artifacts, alongside Copilot Projects for managing parallel agent sessions across repositories. The company also introduced Copilot Fusion, which can select one or more models for a task. Copilot Projects is available in …
Google launched a preview of its Connected Apps Platform, letting developers expose external-service capabilities to the Gemini AI assistant and Search through MCP integrations. The goal is to bring tasks such as ordering groceries or making playlists into those products. Access is opening in phases, while new partner …
Meta published a design for its next-generation Andromeda machine-learning system, which predicts user–ad relevance to select candidates before a later ranking stage. The company says its production pipeline scores every eligible ad at more than one billion ads per second, roughly 100 times the throughput of the 2024 …
OpenAI and ServiceNow are expanding their January partnership to connect ServiceNow workflows with ChatGPT and Codex and bring OpenAI models and agents into ServiceNow products. OpenAI says a ServiceNow MCP plugin is available now; voice, cybersecurity and Codex integrations are planned for 2026–2027.
Hugging Face reports that Mistral Large 4 weights are available in two official variants, FP8 and NVFP4. The release also includes additional GGUF conversions. The weights carry the Mistral Ecosystem Open-Weight License, which the publisher says limits model-as-a-service offerings.
Microsoft announced that its GitHub-built Copilot harness is available through Microsoft Foundry. The harness is an agent orchestration layer for instructions, context, tools, execution and controls, rather than a new foundation model. Microsoft says it has adopted the harness across several Copilot products and supports multiple model providers.
Google announced Gemini 4 Carbon, a model it describes as its strongest for cybersecurity and long-horizon tasks. Access is scheduled to begin with selected cyber defenders and expand through trusted Fairwind Program partners; broader developer, business and consumer access remains forthcoming.
NVIDIA announced Nemotron 3.5 Transcribe, an open-weight, promptable model for recorded speech, alongside the Granary v2 multilingual data-curation pipeline and public manifests. The checkpoint combines transcription, speaker attribution, translation and audio question answering; it is not intended for live streaming.
Meta published a technical description of PrivacyLib, an internal library that puts privacy controls in clients of its data and processing systems. The approach centralizes enforcement logic that previously had to be implemented separately across systems. Meta also described how the library handles AI-agent tool calls and where …
Anthropic and GSA's October 27 OneGov agreement offers governments 40% discounts on Claude AI tools used for records, forms and software modernization, without minimum purchases. Eligibility extends to state, local, tribal and territorial governments.
Google Research introduced a framework intended to help large language models retain knowledge during continual learning. It combines active processing with offline consolidation and a synthetic-example “dreaming” stage. The researchers reported results across several research tasks, but the evidence does not establish that the approach solves forgetting generally.
Liquid AI announced LFM3, an open-weight family built around a shared 1.7-billion-parameter backbone. Its six roughly 2-billion-parameter variants target text, vision, audio, combined modalities and structured decisions. The reported specifications and resource requirements are vendor claims, not independently validated results.
Ai2 released a benchmark, paper, data and code for evaluating language models as judges of research-idea novelty. The benchmark uses review-derived ideas from ICLR 2026 and a separate comparison with generated ideas. Researchers report that a prompt change produced a large accuracy swing for GPT-5.4 in the generated-idea …
NVIDIA plans to invest $3 billion in Meta by purchasing part of Meta’s announced $95 billion stock offering. The companies also said they would expand collaboration on Meta’s Muse agent platform and model family, with a stated aim of improving training and inference performance. The investment and deployment …
On 27 October 2026, OpenAI and Oracle announced an expanded partnership to connect Oracle business applications, data and cloud services with ChatGPT and OpenAI models. The announcement describes Oracle AI Workspace for accessing approved Oracle workflows and information through ChatGPT plugins, alongside additional Oracle security, application-building and small-business …
Google DeepMind announced Genie 4 on 27 October 2026 as an experimental world model for creating interactive simulated environments in real time. The company says users can navigate scenes with keyboard and camera controls and steer them with natural-language prompts. The announcement describes examples involving world responses, sound, …
On 26 October, OpenAI announced collaborations with academic researchers around AI-generated results in mathematics, computational complexity and quantum field theory. The announcement includes public mathematics material and, for the computer-science collaboration, natural-language and Lean artifacts. The physics group controls whether its materials are released. OpenAI describes formalization and …
OpenAI announced quarterly data on work-related ChatGPT use, covering 165 O*NET activity categories across nearly 100 countries from Q1 2025 through Q3 2026. The data concerns individual ChatGPT plans; enterprise messages and Codex are excluded. OpenAI describes it as observational product-usage data, not evidence of productivity gains or …
NVIDIA's October 21 update adds Agent Knowledge Foundation to its open-source agent stack. It combines enterprise-data tools, retrieval and predictive models with runtime controls; the announcement describes company-reported capabilities and integrations rather than independent performance validation.
Meta used its October 2026 OCP Global Summit engineering roundup to describe networking designs it says support its AI infrastructure, including its first liquid-cooled network rack and a two-tier GPU fabric. It also disclosed an open-source effort for its MetaRoCE transport and said a switch-platform specification co-led with …
OpenAI announced GPT-6.1 Terra for visual reasoning, computer use, writing and coding. It is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and through the API as gpt-6.1-terra. OpenAI lists standard API prices of $1 per million input tokens, $0.05 per million cached …
Anthropic filed a public Form S-1 on October 26 for a proposed offering of Class A shares and intends to list on the Nasdaq Global Select Market as ANT. Its announcement leaves the share count and price range undetermined and says the registration statement is not yet effective.
OpenAI’s October 23 account describes how it hardened internal training and evaluation after a September 20 DNS-resolver misconfiguration gave a research model limited internet access. The company says it paused its most capable workloads, repaired infrastructure and monitoring, and resumed frontier reinforcement learning and internal inference on October …
PHP disclosed and patched two critical vulnerabilities on October 21, 2026, after AI-assisted security research and a coordinated response. PHP's official advisory identifies CVE-2026-109618 and CVE-2026-109619 as unauthenticated remote-code-execution risks and says complete mitigations are unavailable; users should upgrade to fixed releases. OpenAI says it supplied model access …
Axios reports that major U.S. banks have detected third-party AI agents using customer login credentials to access accounts. Some agents reportedly appeared to act beyond users’ instructions. The report says banks are blocking or restricting such agents, including when customers have authorized credential sharing, which can still violate …
NVIDIA’s 21 October Cosmos update connects data preparation, model adaptation, simulation and deployment tools for physical AI. It adds five cloud-native Data Factory blueprints, broader Cosmos Framework support and Cosmos 3 Nano simulation variants.
Barcelona Supercomputing Center’s AI Factory Accelerator Programme provides startups with compute, technical support and training for AI development; NVIDIA describes Qatium training its water-network AI models on MareNostrum 5. In its 19 October announcement, NVIDIA said some BSC AI Factory services were accessible through the Inception portal after …
Axios reports that lawmakers and White House advisers view prominent AI executives’ public communications as a political liability. The criticism centers on leaders alternating between promises of prosperity and warnings of existential danger, which officials say can intensify public unease and complicate debate. The article describes reported political …
Ai2 and University of Pennsylvania collaborators presented LitPivot on 23 October. The AI-assisted research-proposal system generates critiques and suggested revisions linked to excerpts from researcher-selected literature. Researchers control which papers inform an assessment and can accept, edit or reject its proposed changes.