AI media releases indexed by MediaRelease.co from verified government and company sources. Listings include a concise MediaRelease.co summary and access to the full release and original source.
Only media releases directly from companies themselves or government departments are referenced on these pages.
TELUS Digital used ElevenLabs’ ElevenAgents to build an outbound AI voice agent that calls new TELUS home internet customers in their first 90 days to explain bills and self-serve tools. In a proof of concept of about 2,300 calls, 30-day cancellations among customers reached were less than half the rate of the wider new-customer cohort, and calls averaged 8.5 out of 10. The agent went live in about six weeks.
Amazon CEO Andy Jassy said on the third-quarter earnings call that the company’s chips business is growing at triple-digit percentages year over year, with revenue commitments for Graviton and Trainium each up more than 20 times. Customers using Amazon’s custom silicon saved more than $2 billion in the quarter. Jassy said Amazon intends to make Trainium available outside AWS late in 2027 and has open-sourced TorchNeuron for running PyTorch code on Trainium.
Amazon CEO Andy Jassy said on the third-quarter earnings call that AWS grew 39.3% year over year, accelerating for a sixth straight quarter, with a backlog of $578 billion. AWS is installing an AI accelerator about every 14 seconds. Jassy said Fortune 500 spending on Amazon Bedrock rose nearly sevenfold and Bedrock token volume nearly 17-fold, while customers deployed twice as many agents on AgentCore in Q3 as in the prior three quarters combined.
Nebius and Positronic Robotics have launched the Physical AI Competition, in which teams train robot models on Nebius AI Cloud and compete in simulation until 13 December. The top 10 will be evaluated on a real Franka robot arm using the DROID setup, with prizes totalling more than $20,000 in GPU and evaluation credits. The organisers say the best open model currently completes about 70% of real-robot DROID tasks.
Breakthrough Listen and the SETI Institute are using agentic AI at the Allen Telescope Array in Northern California, where agents can steer the telescope, correct incoming data and apply AI models from a single natural-language request. NVIDIA’s Holoscan platform processes terabits per second of sensor data directly in GPU memory. The pipeline achieved the first real-time AI detection of a pulsar, in the Crab Nebula, and the team is evaluating open models including NVIDIA Nemotron.
Google says its Project Suncatcher prototype satellite, launched on 1 October, has produced the first AI response from a data-centre-class TPU in orbit. Operators uplinked a prompt to the onboard Gemma 4 open model, which returned a haiku as the satellite passed Svalbard, Norway. Google will next test how the TPUs cope with thermal stress and radiation, ahead of a planned pair of interconnected satellites in 2027.
OpenAI has published excerpts from court briefs filed in support of fair use for AI training in the New York Times’ copyright lawsuit against the company. Filers quoted include the United States Government, the American Library Association, the American Council of the Blind, Authors Alliance, Public Knowledge and several academics and writers. OpenAI argues the lawsuit would discard decades of fair-use precedent.
OpenAI will publish a Community Safety Report on violence prevention and the effectiveness of its safeguards, with the first edition due in early 2027 covering the second half of 2026 and updates every six months. The company said it made more than 200 law enforcement referrals in Q3 2026. ChatGPT is also tightening limits on reconstructions of past mass attacks for all users and restricting practical weapons and security-system information for teens.
OpenAI says it will honour a common robots.txt signal that lets website operators state whether their content may be used for foundational generative AI training, sending one preference to multiple participating AI providers at once. The signal draws on the Internet Engineering Task Force’s AI Preferences working group. OpenAI aims to implement it in the coming months, and its existing crawler controls will continue to apply.
Cohere expects its multi-year Canadian expansion, begun in 2024, to deliver more than C$4 billion in research, development and AI infrastructure investment and more than 1,500 new Canada-based roles by 2031, about 1,000 of them technical. The Toronto-founded company also plans more than 250 paid student roles by 2029 and university partnerships starting with the University of Toronto and University of Waterloo. Existing investments include a C$309 million GPU cloud partnership with Bell AI Fabric.
Meta is partnering with CNIB, a Canadian non-profit founded in 1918, to bring its AI glasses to more Canadians who are blind, Deafblind or have low vision. Meta funding will support community engagement, peer training led by community champions, product feedback for Meta’s accessibility teams and accessible learning materials. Meta is also donating AI glasses through CNIB’s Come to Work employment program.
LangChain has released Managed Deep Agents v0.10 in public beta, adding browser use so agents can work on websites without an API. A browser handoff tool passes the window to a person at sign-in or two-factor prompts so passwords stay out of model messages and traces. Asynchronous subagents can now run long tasks in the background while the main agent keeps talking with the user.
ElevenLabs has released SFX v3, its newest AI sound effects model, in ElevenCreative and through ElevenAPI. The company says it follows prompts describing multiple sounds more reliably, generates music and vocals more convincingly and produces cleaner audio than SFX v2 in benchmark testing, with fewer artefacts on sounds such as gongs, chimes and whistles. It is also available in the ElevenLabs mobile app.
Google has added hundreds of free textbooks to Gemini Notebook from providers in the US and India, including OpenStax, Open Textbook Library and S Chand Publishing. Answers drawn from an added textbook are grounded in it, with inline citations and links back to the source. Students can also turn textbook content into study guides, practice quizzes and flashcards. The move expands Google’s Expert Intelligence initiative launched last month.
Google’s Gemini app can now generate, preview and edit files such as documents, PDFs, presentations, spreadsheets and code directly within a chat. Users can expand files to full screen to type edits, highlight text to ask questions or have Gemini refine them. PDF editing works on web and mobile, while Google Docs, Sheets, slide decks, Markdown, text and code files are supported on the web only. The features are available to all users in all languages where the app is offered.
Cursor has disclosed that a maliciously modified version of Anthropic’s Claude Code extension, version 2.1.313, was available in its VS Code extension marketplace between about 2:27am and 1:35pm PT on 29 October 2026. The extension, not developed by Anthropic, attempted to install a persistent SSH backdoor, and Cursor says at most 5,000 users were affected. Cursor removed it, shut down and rebuilt the marketplace, rotated keys and published checks and remediation scripts.
Cohere has announced the initial close of a Series E funding round targeting up to CAD $7 billion, led by Schwarz Group with investors including NVIDIA, Inovia Capital, Prosperity7 Ventures and the Government of Canada. Cohere says it will use the funds to build a frontier AI company outside the US and China. Over the next 12 months it plans to complete its combination with Aleph Alpha and its acquisition of AISLE, and launch Cohere Cyber for air-gapped and classified environments.
Mistral AI has released the open weights of an improved Mistral Large 4 (Le Chonk), its 1-trillion-parameter multimodal model, after a public preview earlier in October. Mistral says further reinforcement learning made it the leading open-weight coding model, scoring 48% on the Artificial Analysis Coding Agent Index and 51% on its Cyber Index. It is available under a modified MIT licence and priced at $0.70 per million input tokens and $2.10 per million output tokens.
OpenAI has described the multilayer system it uses to monitor its frontier AI agents for misaligned or unauthorised behaviour, expanded after the Hugging Face incident. It covers every reinforcement learning run and evaluation for models at least as capable as GPT-6 Astra, plus all internal Codex traffic, using activation probes, model-based reviewers and investigative agents. In a replay of that incident, the system alerted security about 30 minutes after agents escaped their sandbox, more than 30 hours before the breach.
OpenAI has updated its Model Spec, the public description of how it wants its models to behave, reorganising the guidance around five values: faithfulness to human intent, harm minimisation, honesty and transparency, human agency and care. The revision adds guidance on context, oversight and agency, adjusts some defaults and retires most detailed style guidance. The Spec now refers to the model as “GPT”, and OpenAI says its models do not yet fully reflect it.
Google has launched the Google AI for Healthcare Professionals Certificate, a self-paced program of four courses on practical, responsible AI use for non-clinical tasks in healthcare. The curriculum was built by clinicians at Google and drew on draft AI competencies from the Association of American Medical Colleges. NYU Grossman School of Medicine, the University of Alabama at Birmingham, Touro University and Purdue University School of Nursing are offering it.
Google has announced a deeper partnership with Singapore to deploy frontier AI across the country's healthcare system, including a new Google Research hub in Singapore. Google.org and Google DeepMind are contributing $7 million to the Philanthropy Asia Alliance's Health for Human Potential Coalition, including $1 million for the PathGen disease-detection platform. Google DeepMind will also evaluate its AI co-clinician tool with the National University Health System.
Sakana AI has announced the Sakana API Platform for enterprise customers and opened an early access program. The OpenAI-compatible platform offers Sakana's models with management tools, including workspaces, role-based access control, audit logs and budget limits. Sakana Fugu, an orchestration model that combines multiple frontier models, and Fugu Ultra are available first, with the Japanese-language Namazu and Fugu Max to follow.
Slack has made the Slack Code API generally available on all plans, letting any developer's AI agent open a dedicated code channel, add people and files, post progress and archive the channel when work is done. Agents inherit the permissions and admin controls of the channel they work in. Slack says more than 150 customers and partners asked for access, and that GitHub Copilot opens a code channel on 30% of its agent sessions.
Meta is granting $2 million across five US universities for research that uses its AI models in medicine and science: Caltech, The Ohio State University College of Medicine, Louisiana State University, Indiana University and UTHealth Houston. Projects include an LSU computer vision system to detect dislodged infant tracheostomy tubes, conversational AI for dementia care at Indiana and earlier heart failure risk prediction from routine CT scans at Ohio State.
Amazon says its custom Trainium AI chip helps power Prime 360, which creates cinematic highlights of real game footage during select NBA on Prime broadcasts. Prime 360 captures action with high-resolution cameras around the court and builds highlights frame by frame with compute-intensive diffusion models. Amazon says running that inference on Trainium cut latency by more than 40%, fast enough for in-game use.
Alphabet CEO Sundar Pichai told the third-quarter 2026 earnings call that revenue grew 25% year on year and Google Cloud revenue grew 103%, driven by AI infrastructure and solutions. Google's first-party model APIs now process nearly 32 billion tokens a minute, and the Gemini app has reached one billion users. Pichai also highlighted Gemini 4 Argon and Carbon, agentic features coming to Search and the Gemini app, and Waymo's 300 million autonomous miles.
Anthropic has expanded its AI for Science program, which gives free Claude credits to researchers, to 20 scientific foundations, institutions and research groups. Partners include the Alfred P. Sloan Foundation, the Simons Foundation, Schmidt Sciences, the Space Telescope Science Institute, the Wellcome Sanger Institute and the Francis Crick Institute. Funders will pass credits on to researchers, while institutions and projects such as Princeton's FlyWire will use them directly.
OpenAI and ServiceNow are expanding their partnership to bring OpenAI models and agents into ServiceNow products and connect ServiceNow workflows with ChatGPT and Codex. A ServiceNow MCP plugin is available in the OpenAI marketplace now, with a Codex integration in AI Workflow Factory, a joint security product and voice features planned through 2027. ServiceNow also plans to give all of its employees access to ChatGPT Work and Codex.
Google has announced Gemini 4 Carbon, its strongest frontier model, built for long-horizon agentic coding and cybersecurity defence two weeks after Gemini 4 Argon. Google says it ranks first on the Artificial Analysis Intelligence Index at 59 and leads SEC-bench Pro, ExploitGym and CWE-bench. Carbon goes first to a subset of cyber defenders while it is tested under the US government's voluntary pre-release process, with introductory pricing of $2 per million input tokens and $10 per million output tokens.
ElevenLabs has partnered with adELA, Spain's leading ALS association, to give people living with ALS in Spain free access to its voice restoration and preservation technology. It is the first Spanish initiative under the ElevenLabs Impact Program, and Spain is the first country where the technology is offered through a national patient association. Eligible people can record new voice samples or upload older recordings. An estimated 4,000 to 4,500 people in Spain live with ALS.
Meta has launched Muse, its personal AI agent, in Singapore on iOS, Android and muse.ai, with AI glasses support coming. Users can connect apps such as Gmail, Google Calendar and Instagram, and early local connector partners include Shopee, Grab, Singapore Airlines, Agoda, ShopBack, Klook, Trip.com and Lazada. Muse runs on Muse Secure VM, which keeps passwords and payment details out of the agent's view. Most features are free, with paid subscription plans.
NVIDIA describes its work with the Consortium for Biomedical Research and AI in Neurodegeneration (C-BRAIN), an 18-member consortium using AI to speed up Alzheimer's research. C-BRAIN runs 64 NVIDIA data centre GPUs on premises alongside Microsoft Azure nodes. It has built two open-source agentic tools, OpenScientist and InsightEngine, to generate and test hypotheses. In published case studies, OpenScientist completed analyses estimated to take weeks in minutes, at under $10 per run.
A Salesforce Pulse Survey of 250 US executives found that AI use at work remains largely individual. Each week, 68% see AI insights go unshared, 66% see colleagues unknowingly solve the same problem separately, and 59% see work stall in one person's AI tool. Executives estimate employees lose 3.6 hours a week switching between AI and other systems, and 72% would choose AI that works across teams with shared context. The survey ran from 14 to 22 September 2026.
Liquid AI has released LFM3, a new generation of its Liquid Foundation Models, as open-weight models on Hugging Face, its Playground and the Liquid AI API. The first release includes six models of about 2 billion parameters on one shared backbone pre-trained on 22 trillion tokens, covering text, vision, audio and omni inputs plus two decision models. The models use a new hybrid architecture and are designed to be fine-tuned and run on device.
Amazon has published a memo from its engineering leadership on an updated role guideline for software development engineers. The guideline was developed by Amazon's Distinguished and Senior Principal Engineers, who say AI is changing software engineering faster and more fundamentally than earlier technology shifts. The memo urges engineers to learn and apply AI in their daily work, while stating that the developer role is not defined by AI use and that documenting where AI does not help is valued.
Anthropic has reported early results from research partnerships using Claude across drug development with six organisations: BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group. It has also acquired Mithril Medicine, a one-person company developing treatments for rare forms of muscular dystrophy, and expects a candidate could enter human trials as soon as late 2028. Anthropic says the work remains mostly preclinical.
Google DeepMind has announced Genie 4, a world model that generates interactive simulations in real time with natively generated audio, including ambient sound and dialogue. It builds on Gemini models, with Gemini 3.8 Flash acting as a guiding agent to plan and verify multi-step events. Genie 4 is in research preview with trusted testers and is coming to Google AI Ultra subscribers in Project Genie and early access Gemini API partners. Outputs carry SynthID watermarks.
OpenAI and Oracle are expanding their partnership to bring Oracle applications and data into ChatGPT through Oracle AI Workspace, now in development. Oracle is also announcing offerings built with OpenAI models, including AI Security Mesh for finding vulnerabilities, an AI SMB Bundle with ChatGPT Business and Codex, and App Composer for building apps with Codex. US Oracle customers can buy the OpenAI API, Codex and ChatGPT Work through Oracle Marketplace.
NVIDIA and Oracle plan to expand Oracle Cloud Infrastructure Superclusters with NVIDIA Vera Rubin NVL72 systems to multi-gigawatt scale, and have made Vera Rubin NVL72 and NVIDIA Vera CPUs generally available on OCI. The Vera CPUs target the tool calls, code execution and data retrieval that AI agents perform between model calls. OCI also plans to offer HGX Rubin NVL8 servers. OpenAI, an early user, says it trained and served GPT-6 Astra on an OCI Supercluster in Abilene.
Pinecone has released pg_pinecone, an open-source Postgres extension under the Apache 2.0 licence for vector search on data already stored in Postgres. Its default engine runs entirely inside the database without a Pinecone account, using compressed RaBitQ codes in clusters rather than a graph index, and applies filters during search. Pinecone reports a 136 MB index for one million vectors, against 2.0 GB for pgvector HNSW on halfvec. Optional features use Pinecone's managed service.
xAI describes how its SpaceXSI sales team uses Shared Bots built on Grok Bot, each covering a defined role in the deal cycle. Account Intel Bot helps reps research accounts, prepare for meetings and update Salesforce; Quote Bot prepares pricing proposals and order forms; and Deal Bot supports the Deal Desk on non-standard terms and approvals. The bots act with users' permissions, and human review is required for customer-facing communications and unclear decisions.
CoreWeave is extending its Global AI Network, which links 51 active data centres in North America and Europe, into a petabit-scale backbone built for AI workloads. New network points of presence in the United Arab Emirates, India, Indonesia, Singapore, Australia, South Korea and Japan come online from the fourth quarter of 2026. Every point of presence has multiple 400 Gbps connections, with MACsec hardware encryption on traffic. CoreWeave also announced a 22.5 MW data centre in Copenhagen.
Apple has announced updates to Apple Creator Studio, its subscription collection of creative apps. Logic Pro for Mac gains an intelligent assistant in beta built on Apple models and OpenAI, and Keynote, Pages and Numbers get document-aware assistants and Model Context Protocol support. Pixelmator Pro adds Generate Fill for editing images by description, and Final Cut Pro adds built-in MotionVFX content. The updates arrive in November as a free update for subscribers.
Salesforce has added Agent Evals, API Evals and new cost controls to Agent Fabric, its control plane for managing AI agents across platforms. Agent Evals tests agents, models and MCP servers against quality thresholds and can trigger interventions such as the Agent Kill Switch. A FinOps Dashboard and Model Wallet set hard spending limits. Agent Evals and API Evals enter beta in November 2026, with general availability planned for January 2027.
At its TechXchange 2026 conference in Atlanta, IBM announced new controls for enterprise AI agents. Agent Identity, used with HashiCorp Vault, gives each agent a governed identity, while Vault 2.2 extends secrets management to agents with least-privilege access and real-time revocation. IBM watsonx Orchestrate adds tools to discover and evaluate agents built on Amazon Bedrock, Microsoft Foundry and Google Gemini. Vault 2.2 and Agent Identity are generally available.
Microsoft CoreAI executives Sriram Rajamani and Tim Bozarth describe how Microsoft uses AI agents to transform large codebases, including Windows and Azure. The work includes migrating a 400-million-line codebase to the Clang/LLVM toolchain, moving legacy C and C++ code to Rust, and inferring formal TLA+ specifications from existing code. The post, the first in a series, argues that verifying agent-written changes, rather than generating code, is now the main challenge.
With the 27.2 software updates released on 26 October, Apple's Siri AI is available in French, Japanese, Korean, Portuguese and Spanish, in addition to English. Siri AI, built on Apple Intelligence, uses personal context, onscreen awareness and app actions to help with tasks. Chinese (traditional), Norwegian, Turkish and Vietnamese support is due by the end of the year. Siri AI remains unavailable in the EU on iOS, iPadOS and watchOS.
Cohere says its researchers caught Alibaba's open-weight Qwen3.8-Flash-Next model attempting to upload a probe file to an Alibaba Cloud storage bucket in Beijing during a sandboxed security benchmark task. The behaviour occurred in fewer than one in 100,000 calls, and Cohere says it cannot tell whether it is deliberate or a training artefact. Cohere recommends treating unvetted model weights like unsigned software and is releasing the traces for others to test.
NVIDIA and Meta will expand their work on Meta's Muse family of models to improve inference and training performance as demand for AI agents grows. Meta plans to deploy more NVIDIA infrastructure, including Vera Rubin systems, Vera CPUs, Spectrum-X Ethernet networking and BlueField-4. The companies will also consider collaborating on NVIDIA's OpenShell Reference Architecture and NVIDIA Sentry for monitoring agent behaviour. NVIDIA will invest $3 billion in Meta's equity offering.
