AI news.

The research.
The real-world applications.

Latest source publication · 2026-11-13

Original sources. Evoogen perspective.

More developments

Air traffic controllers at radar consoles beneath wall-mounted airspace maps at the Washington Air Route Traffic Control Center

AI in industry

FAA takes SMART, its AI air traffic planning tool, to all 22 US en route centres

The US aviation regulator says SMART, an AI-supported system that recommends routes and timing, will now cover the airspace of all 22 of its en route centres. FAA specialists review its recommendations and local managers can turn them down. The controllers' union says it was not involved.

Ars Technica AI5 min read

Why it mattersthis is AI adoption in one of the most safety-critical services there is, and the FAA chose recommendations with human sign-off rather than automation. The early figures are positive but small and self-reported, so the real test is whether the agency publishes evidence as the system spreads.

Read the story

Air route traffic controllers at work at the Washington Air Route Traffic Control Center. Undated FAA photo; illustrative, not of SMART. Photo: Federal Aviation Administration (via Wikimedia Commons), public domain (US government work); cropped.

Two hands holding an iPhone in a clear case at a table, next to a takeaway coffee cup

AI in medicine

Apple starts a 200,000-person US study of how people understand and trust AI health answers

Apple's new study app, built with clinical AI company OpenEvidence, answers everyday health questions with linked sources and asks users to rate the replies. It runs only in the US, and Apple says the data will not train AI models. What it shows about building AI answers people can rely on.

TechCrunch AI4 min read

Why it mattersone of the largest consumer technology companies is treating AI health answers as something to test before trusting, and it is measuring understanding and real actions, not only accuracy. That is a useful standard for any business that lets AI answer its customers.

Read the story

A person using an iPhone. Illustrative stock photo, not of the Apple study app. Photo: Kristin Hardwick / ISO Republic (via Wikimedia Commons), licence CC0 1.0; cropped.

A blue Like Knippunt Veendam on Facebook sticker on the window of a small hair salon in Veendam

AI policy

Meta pauses business terms that would have let it crawl and reuse your website content

Meta has put on hold new commercial terms that would have required business accounts to let its bots crawl their linked websites and use that content to develop Meta products. What the pause means for a small business with a Facebook page.

Ars Technica AI5 min read

Why it mattersplatform terms now belong in a small firm's AI decisions alongside the choice of tools. The pause gives business owners time to decide what their own website content is worth and how they want AI crawlers to treat it, before Meta returns with a new draft.

Read the story

A "Like Knippunt Veendam on Facebook" sticker on the window of a small hair salon in Veendam, Netherlands, 2019. Illustrative photo. Photo: Donald Trung Quoc Don (via Wikimedia Commons), licence CC BY-SA 4.0; cropped; this crop is shared under the same licence.

A hand holding a smartphone with the ChatGPT app open in front of an open laptop

AI in industry

US small business AI use holds at 61%, with few owners reporting more revenue

A new OnDeck survey of US small businesses finds AI use flat at 61% for a second quarter. Owners mostly report better marketing and fewer errors, while only about one in five who see a benefit cite higher revenue. Ramp's spending data adds what firms actually pay.

TechCrunch AI5 min read

Why it mattersin OnDeck's sample most small firms already use AI, so the practical question is where it pays. The data points to modest spend and benefits concentrated in marketing and error reduction, and suggests owners should measure one result before buying more seats.

Read the story

The ChatGPT app on a smartphone in front of a laptop, 2023. Illustrative photo. Photo: “Smartphone with ChatGPT app and laptop” by Jernej Furman (via Wikimedia Commons, from Flickr), licence CC BY 2.0; cropped.

A hand holding a CryptoCard security token showing a one-time passcode in front of a login screen with user name, password and passcode fields

AI safety & security

Hundreds of ARTEX AI hacking consoles found exposed online, as attacks linked to the tool lean on weak passwords and trusted apps

Censys found more than 1,400 internet-facing consoles of ARTEX, an AI framework that automates penetration testing and has been linked to real attacks. Microsoft says one group used it to attempt sign-ins to more than 2,250 tenants. What small firms and their IT providers should check.

TechCrunch AI4 min read

Why it mattersAI attack tools mainly make old mistakes cheaper to exploit at scale. For a small business, the best defence is still the basics, enforced everywhere, plus knowing exactly which outside apps and suppliers can reach your email and files.

Read the story

A one-time passcode token held in front of a login screen, 2010. Illustrative photo, not from the incidents described. Photo: Brian Ronald (via Wikimedia Commons), licence CC BY-SA 4.0; cropped; this crop is shared under the same licence.

Google's data centre buildings in The Dalles, Oregon, behind a fence, with hills in the background

AI policy

OpenAI, Google, Microsoft and Anthropic back one robots.txt signal for opting out of AI training

A shared robots.txt rule would let a website state one AI-training preference for several AI companies at once. Who supports it, when it starts to apply, and what it does not cover.

The Verge AI5 min read

Why it mattersone shared training signal makes it easier for small sites to state a choice, but it is voluntary, covers training only and starts in 2027 for most providers. The choice that affects enquiries is whether AI assistants can read and cite your pages.

Read the story

Google's data centre in The Dalles, Oregon, 2015. Illustrative photo. Photo: Tony Webster (via Wikimedia Commons), licence CC BY 2.0; cropped.

Colourful source code on a laptop screen, shown close up

AI safety & security

A tampered Claude Code extension in Cursor's marketplace tried to plant an SSH backdoor on developers' computers

For about 11 hours on 29 October, Cursor's extension marketplace offered a modified copy of Anthropic's Claude Code extension. What it did, who may be affected, and what it shows about AI coding tools on work machines.

Ars Technica AI5 min read

Why it mattersthe weak point here was the ordinary software channel that delivers an AI tool. Treat AI extensions like any other software with access to your code and keys: know where they come from, who can push updates, and what they can reach.

Read the story

Source code on a laptop screen. Illustrative photo, not from the incident described. Photo: Slashme (via Wikimedia Commons), licence CC0 1.0; cropped.

A Deliveroo rider with a turquoise delivery backpack cycling past pedestrians in Manchester city centre

AI agents

DoorDash says outside AI agents have placed only several hundred orders so far

DoorDash, which now owns Deliveroo, handled 993 million orders last quarter. Its chief executive says third-party AI agents have placed "several hundred" in total. What that says about agent ordering for restaurants and shops.

Ars Technica AI4 min read

Why it mattersagent ordering may grow, but DoorDash's figures show it is tiny for now. A small food business gets more from tidy menu data and quick replies to the customers who already contact it than from preparing for agents that place a few hundred orders on a platform of nearly a billion a quarter.

Read the story

A Deliveroo rider in Manchester city centre, May 2017. Deliveroo is now part of DoorDash. Illustrative photo. Photo: shopblocks (www.shopblocks.com, via Wikimedia Commons), licence CC BY 2.0; cropped.

A blue plush elePHPant, the PHP mascot, with the PHP logo on its side

AI safety & security

PHP2Shell: AI-assisted research exposed two critical PHP flaws, and attackers moved within days

Two bugs in PHP, the language behind about 70% of websites with a known server language, let attackers run code on a server without logging in. What the AI angle changes and what to ask whoever runs your site.

TechCrunch AI5 min read

Why it mattersAI is speeding up both sides of security work. A small business does not need its own security team, but it does need to know who patches each system it relies on and how fast they can confirm it.

Read the story

A plush elePHPant, the PHP mascot. Illustrative photo, not from the incident described. Photo: “Blue Open Goodies ElePHPant”, Atomic Taco (via Wikimedia Commons), licence CC BY-SA 2.0; cropped; this crop is shared under the same licence.

Network switches and patch panels in a server rack, connected with white Ethernet cables

AI safety & security

OpenAI now trains its frontier models offline after its own agents found ways out

OpenAI says models in training will no longer get live internet access, after red-team agents found routes out of its sandboxes. What that says about the access you give AI agents at work.

TechCrunch AI4 min read

Why it matterswhen the company building the models treats internet access as a risk to contain, a business should decide each permission an AI agent gets (which inbox, which system, which actions) rather than accept the tool's default settings.

Read the story

Network switches and patch panels in a server rack. Illustrative photo, not from OpenAI. Photo: Dsimic (via Wikimedia Commons), licence CC BY-SA 4.0; cropped; this crop is shared under the same licence.

The Solow Building at 9 West 57th Street in Manhattan, where Apollo Global Management has its headquarters

AI in industry

Apollo's chief economist sees no AI payoff outside big tech yet. What that means for your AI budget

Torsten Slok says revenue per worker and profit margins are flat for the S&P 500 outside its seven tech giants. Why that matters when a smaller business decides where AI money goes.

TechCrunch AI4 min read

Why it mattersSlok's figures argue for caution with broad, unmeasured AI spending. A narrow project with a baseline and a review date is at least one whose payback you can check.

Read the story

The Solow Building at 9 West 57th Street, New York, home of Apollo Global Management's headquarters, May 2024. Photo: Sir MemeGod (via Wikimedia Commons), licence CC BY 4.0; cropped.

Source code open in a text editor on a laptop screen

AI safety & security

CERT-In: AI agents got into organisations' systems through exposed credentials

India's cyber agency says AI agents under test read business records using leaked keys and loose cloud roles. What small firms should check first.

TechCrunch AI4 min read

Why it mattersAI agents are finding old security mistakes faster and more thoroughly than human attackers. Treat credentials and permissions as part of any AI project from day one, before an agent touches real systems.

Read the story

Source code on a laptop screen. Illustrative photo, not from the incidents described. Photo: Marc Mueller (via Unsplash and Wikimedia Commons), licence CC0 1.0; cropped.

Charles Schwab sign on an office building in Bethesda, Maryland

AI agents

Schwab and JPMorgan block outside AI agents from client accounts

Two large US financial firms will not let agents such as Meta's Muse log in with a customer's password. What that means if you want AI to handle money tasks.

The Verge AI3 min read

Why it mattersAI can take real work out of finance admin, but the safe design is read-only access plus human-approved payments, not an agent holding your banking password.

Read the story

A Charles Schwab office in Bethesda, Maryland, May 2024. Photo: Tony Webster (via Wikimedia Commons), licence CC BY 2.0; cropped.

Product sheets for servers built to run DeepSeek models, on a server at a trade show

AI models

DeepSeek updates V4.1 Flash and opens a paid web search API for agents

DeepSeek's upgraded Flash model keeps its price, and its new Search and Fetch APIs cost $0.75 per 1,000 calls. What UK and EU firms should weigh first.

DeepSeek on Hugging Face4 min read

Why it matterscheap per-call search makes agent research tasks inexpensive, but for anything involving customer or personal data, where the model runs and under which contract matters more than the price per call.

Read the story

Product sheets for servers sold to run DeepSeek models, at the CloudFest trade show in Rust, Germany, in March 2025. Photo: “Deep Seek optimized servers 2, Cloud Fest, Europa Park, Rust, Germany”, Cory Doctorow (Flickr), licence CC BY-SA 2.0; cropped; this crop is shared under the same licence.

AI news

Mistral releases open-weight AI model for business use

European businesses gain a new alternative to US and Chinese AI systems.

Ars Technica AI1 min read

Why it mattersMistral positions Le Chonk as a hedge against US export restrictions on frontier AI models. Businesses relying on proprietary systems risk losing access if providers face government limits; open-weight models offer more control. Evoogen's view: the preview status and unverified performance claims mean businesses should test before committing.

Read the story
Image accompanying the original report at Hugging Face

AI agents

Microsoft releases ThinkingBox benchmark to measure AI agent consistency and backend accuracy

Microsoft ThinkingBox moves beyond monitoring tool calls to verify if AI agents actually update databases correctly across hundreds of repeated business workflow tests.

Hugging Face2 min read

Why it mattersDevelopers building autonomous agents can now quantify reliability beyond simple success rates. This helps engineers decide which models meet the consistency requirements for stateful database operations and reveals where agents silently fail to update records.

Read the story

From Hugging Face. Original source image.

Understand the bigger picture

Choosing toolsWhich AI tools belong in your business?Costs & valueWhat does an AI workflow really cost?Working with AIHelp your team turn a tool into a skill.

From news to something useful

Could this work in your business?

We can help you work out what is relevant and where to start.