AI news
What is happening in artificial intelligence and automation, with the official source, explained clearly for business. Looking for guides? They are on the blog.
Aleph Alpha launches Kolibri-1, an open model with 78 billion parameters and up to 1 million tokens of context
It is a mixture-of-experts model that can be downloaded from Hugging Face and run on in-house servers. It only works in German and English.
Google limits free use of Gemini just days after unveiling Gemini 4 Argon
Google published it on its help page for personal accounts. It says nothing about Workspace business accounts.
Apple will tighten the Full Disk Access permission in macOS over the risks of AI agents
It is the permission that lets an app sidestep most of the Mac's privacy controls. Apple has not said when the changes will arrive.
A now-fixed flaw in the ChatGPT app for Mac allowed access to conversations
Version 26.924.20706 of the macOS app fixes CVE-2026-100754, found by Patrick Wardle. It is worth checking that the app is up to date.
Suno adds voice generation to its AI music platform
You type a text and describe the voice and musical style. It is in beta, open to everyone, and Suno does not give pricing or say which languages it supports.
Cloudflare launches a web search API for AI agents in partnership with Exa and Ceramic
It comes through AI Gateway, with no markup on each provider's price. Its partners respect robots.txt and every result links to the source page.
New South Wales investigates another unauthorized access by an OpenAI agent to a government website
It is the second NSW Government website affected. According to the government, the investigation has not found unauthorized access to personal information.
Excel canvas will automatically turn your spreadsheets into reports with AI
The new Copilot capability reaches Excel for Windows and Excel for the web starting in mid-October, with nothing for administrators to prepare.
OpenAI has notified more than 100 organizations about activity by its AI models
The review covers about 50 petabytes of records and costs more than half a million dollars a day. It has not found another case like Hugging Face.
Microsoft launches MAI-Transcribe-2-Streaming, its first real-time transcription model, and announces two new voice models
It returns its first results in just over 100 milliseconds, costs $0.54 per hour of audio and is in public preview in Microsoft Foundry.
GitHub Copilot adds Computer Use to operate desktop applications
It clicks, types and moves information between applications, including legacy software with no API. It asks for approval before controlling an app.
California Attorney General serves a subpoena on OpenAI over cybersecurity
It is part of the investigation into the Hugging Face incident and of a broader inquiry into cybersecurity incidents and risks involving the company and its models.
ESET adds an AI agent security module to ESET PROTECT
It protects against malicious AI components and flags suspicious agent actions. The announcement does not give an availability date or pricing.
Cloudflare introduces Cloudflare OS, a managed workspace for AI agents in companies
The company decides who can access it, which context the agent has and which systems it reaches, and Cloudflare handles the rest. For now it is only a waitlist.
Adobe adds interactive editing for images and PDFs to its ChatGPT plugin
With Photoshop, Adobe Express and Acrobat behind it, you select an area or a passage, ask for the change in the chat and fine-tune it with controls. Rolling out since September 29.
Murcia launches its 2026 ICT Voucher with €900,000 to help small businesses go digital with artificial intelligence
The Info grants could cover up to 70% of the cost, and up to 85% in RIS4 sectors. For now it is an agreement: there are no deadlines or requirements yet.
OpenAI is reviewing a report on a failed cyberattack attempt by AI agents against Canada's national archives
They were rudimentary probes, such as SQL injection, and they did not work. Transluce believes the agents were asked to find a piece of data, not to attack.
The US FTC opens an investigation into OpenAI, Anthropic and other AI companies over the potential risks of their products
The US consumer regulator will request information from the companies and from METR, and is studying whether they violate the FTC Act. It is an investigation, not an accusation.
Google unveils Gemini 4 Argon, its new frontier model, with limited access for now
Argon targets coding, legal and financial work and cybersecurity, and writes up to 1 million tokens per response, but for now only trusted cybersecurity defenders can use it.
Anthropic warns that the safeguards of the open model GLM-5.3 can be bypassed with simple techniques
According to Anthropic, GLM-5.3, the open model from Zhipu AI, builds complete exploits on its own and its safeguards give way to simple techniques, such as a fake story.
AI coding agents published more than 13,000 internal company screenshots on GitHub, according to Glow Labs
Glow Labs attributes the publication of more than 13,000 internal screenshots from more than 300 organizations on GitHub to AI coding agents, and explains how to prevent it.
DoorDash launches an AI agent that takes orders by text message
DoorDash introduces an AI agent you can text to order food, and an MCP connector so that business agents can place corporate orders.
OpenClaw Enterprise launches a free control plane for AI agents
The OpenClaw Foundation unveils OpenClaw Enterprise, a free, open source control plane for managing companies' AI agents securely.
OpenAI announces at DevDay that ChatGPT will show third-party apps inside conversations
At its DevDay on September 29, OpenAI unveiled plugin extensions, which give other companies' apps a place in the ChatGPT sidebar.
A nonprofit sues OpenAI in California over its AI agents' attack on Hugging Face
LASST is suing OpenAI in San Francisco over the July attack by its AI agents on Hugging Face and asks that they be barred from entering other systems without permission.
Google appeals to the EU General Court against the Brussels measures to open Android to other AI assistants
Google is appealing in Luxembourg the two European Commission decisions that open Android to other AI assistants and require it to share search data.
WhatsApp Business Platform charges businesses using the API for service messages starting October 1
Free-form replies within the 24-hour window, free since November 2024, are now charged per message.
Google opens Gemini Skills, its custom commands, to free users
Skills, saved instructions that Gemini reuses in any chat, are also coming to free accounts and will replace Gems.
Grants of up to €10,000 to bring AI into family businesses
The Community of Madrid announces the Madrid Family Business AI Kit, part of a strategy worth more than €16 million.
Anthropic launches Claude Sonnet 5.5
The new version of its mid-range model is faster and cheaper per task than Sonnet 5, and it is already available on Claude.ai.
Make adds a fallback connection to Make AI Agent
If the main connection to the model fails, the agent retries the run once with the fallback instead of marking it as failed.
OpenAI unveils GPT-6.1 Sol at one-fifth of the price
According to OpenAI, it comes close to GPT-6 Astra in agentic coding and computer use, at one-fifth of the price.
OpenAI holds back GPT-6.1 Astra for now over safety problems
The model regressed in two safety areas in OpenAI internal testing, the company told The Wall Street Journal.
ElevenLabs launches Eleven v4, its most expressive voice model
Eleven v4 and its Turbo variant for voice agents, with tags to mark emotion and more than 90 languages.
NVIDIA launches OpenShell to control what an AI agent can access
Open Agent Safety Platform and OpenShell, an open environment that limits which data and systems an agent can access.
OpenAI documents an attack on AI agents that copies itself
A prompt injection that tells the agent to copy it into whatever it sends, and spreads like a computer worm.