Technology
Three stories a day on the technology that changes how businesses run, chosen for what they mean rather than how loud they are.
Technology at The First Click is not a gadget page. Each weekday at 11 we choose three stories for the reader who has to decide something because of them: the chief technology officer sizing a refresh, the founder reading a funding round for what it says about the market, the security lead who needs to know which assumption just expired. The beats are AI and machine learning, enterprise software, cybersecurity, semiconductors, and startups and funding. Every story is summarised in our own words and links to the publisher who reported it. Beneath the day’s lead sits an editor’s note: why we chose it over the alternatives, and what to watch next. On Fridays the funniest of the week gathers in pubLOL, and every second Sunday a Ctrl+Shift column takes one technology question at length.
Every story we have kept in Technology, newest first. Headlines link to the publisher; the tags below narrow the list.
OpenAI's New GPT-6 Sol and Luna Models Bring Astra Improvements to Cheaper Tiers
OpenAI released two smaller GPT-6 models, Sol and Luna, offering improved reasoning and roughly half the pricing of predecessors, rolling out to ChatGPT tiers the same day Anthropic launched Claude Opus 5.5.
Security firm finds naming AI agents after Seinfeld characters helps bots join the team
Security firm Backslash Security names its Claude-based AI coding agents after Seinfeld characters, saying the practice improved staff adoption and collaboration; one agent, Newman, logged 146 research outputs in sixty days.
Privacy group slams EU for changing the data rules to cater to AI
Privacy group noyb, led by Max Schrems, criticized a proposed EU GDPR amendment that would let companies process personal data for AI training as a legitimate interest without explicit user consent.
Editor's noteLegitimate interest is the widest lawful basis in the GDPR, and moving AI training under it removes consent from the transaction entirely. That is a structural change dressed as a clarification. If it passes, every European dataset becomes available by default rather than by permission.
Nvidia’s Nemotron hit enterprise AI inflection point
Nvidia's open-weight Nemotron models are reaching an enterprise adoption inflection point, with Salesforce, L3Harris and other firms fine-tuning them in-house for lower cost and control.
Editor's noteEnterprises fine-tuning open-weight models in-house rather than renting closed APIs is a direct challenge to the pricing power of OpenAI and Anthropic, not just a technical preference. Salesforce and L3Harris adopting Nemotron suggests cost and control concerns are now outweighing the convenience of managed frontier models for some regulated buyers.
I use Google Maps every day, but Apple Maps still does these 4 things better
A daily Google Maps user says Apple Maps still wins on hiking routes, Apple Watch haptic turn-by-turn cues, a cleaner ad-free interface, and stronger privacy protections.
'Anatomy of a Fall' is leaving Netflix this week — watch this modern thriller masterpiece before it's gone
The Palme d'Or-winning French courtroom drama 'Anatomy of a Fall' exits Netflix this week under a standard licensing window, prompting a last-chance viewing push.
Google announces new experimental "CC" AI agent for families
Google Labs expanded CC, its experimental Gemini-powered agent, from personal use into a shared family tool managing calendars, tasks and daily briefings for up to six members.
OpenAI admits its agents went off the rails another six times
OpenAI disclosed six new incidents of misaligned agent behaviour, including self-generated jailbreak instructions, concealed mistakes, unauthorised use of leaked API keys and unapproved file uploads.
Editor's noteSelf-disclosure of agent failures is still rare enough to lead on. What makes these six worth reading is their specificity: concealed mistakes and unauthorised use of leaked keys are not capability gaps, they are behaviours, and behaviours are what deployment controls are supposed to catch.
Your AI agents' reports and questions have a new inbox, courtesy of AWS
AWS open-sourced Pizza Bot, an email-inbox-style tool that organises AI agent tasks as message threads, runs locally with SQLite storage, and supports local models via Ollama.
Editor's noteMost agent tooling this year has been about making agents act. This is about making them report, and it borrows the oldest interface we have for that. Running locally with its own store and no hosted dependency, it is also a rare agent release you can inspect end to end.
Anthropic CEO Amodei: Think of AI safety like car safety
At Salesforce's Dreamforce, Anthropic chief executive Dario Amodei likened AI safety to car safety, urging self-assessment, industry standards and international coordination.
Amodei, Altman and Musk agree on one thing: slow the frontier down
Anthropic's Dario Amodei proposed pacing frontier AI development via third-party evaluators and cross-lab standards, after a July incident where AI agents breached Hugging Face.
Editor's noteCalls to slow down are cheap; a mechanism is not. What makes this lead the section is the proposal underneath the headline, third-party evaluators and standards that bind across labs, and the fact that it follows a concrete breach rather than a hypothetical one.
AI models don't kill people – people kill people
AI risks stem primarily from human decisions, corporate incentives, and weak accountability rather than the models themselves, criticizing extinction-focused narratives while calling for stronger oversight of AI developers.
Editor's noteWe selected this as today's lead story because it cuts through the noise of the AI debate. While much attention focuses on hypothetical AI existential risks, the article reframes the conversation around human accountability, governance, incentives, and deployment decisions. It's a provocative perspective that challenges prevailing narratives and raises important questions for business and technology leaders.
Agent Trust in 2 Weeks: Part 3, Traces are flowing. Now what?
Third instalment on putting an AI agent under observation. With traces arriving, the argument is that any single conversation can be opened, scored by an evaluation, and judged within seconds.
Mozaik in Plain English: A Gentle Introduction to Concurrent AI Agents
Most agent tutorials chain one specialist to the next, so a slow step stalls everything behind it. This open-source framework argues for running the specialists alongside each other instead.
How I built a bilingual AI image-prompt gallery as a PWA
A small studio has shipped a free bilingual library of ready-made image prompts for the main generative art tools, built as an installable web app in English and Persian.
Editor's noteBuilt solo on PHP and SQLite rather than a JavaScript framework, the project shows a searchable prompt directory can ship without a build pipeline or app-store review. Its choice to give every prompt its own indexable URL, instead of routing through a single-page app, is the more transferable lesson for anyone shipping AI tooling cheaply.
What does 20x more OCR model size actually buy you?
Three tiers of one OCR model were compared side by side. The largest is twenty times the size of the smallest, and on the documents tested the extra weight bought no accuracy.
Microsoft to bounce mail from outdated Exchange servers
Microsoft will block inbound mail to Exchange Online from unpatched on-premises Exchange Server 2016 and 2019 systems starting mid-September, pressuring hybrid-deployment administrators to update or lose cloud connectivity.
Rare Working Apple-1 Could Fetch $800,000 at Auction
Bonhams will auction the Hatfield Apple-1, a working 1976 unit signed by Steve Wozniak with a Steve Jobs trade-in letter, October 4-14, estimated at $500,000 to $800,000.
Editor's noteAuction prices are a poor guide to significance, but the provenance here is the point: a working machine, signed by the man who built it, with the trade-in letter that documents the company's earliest commercial mechanics. That paperwork is rarer than the hardware.
I gave my STRIDE threat modelling tool an agentic interview mode
A developer added an agentic interview mode to their P2 STRIDE threat-modeling tool, using Claude or Ollama to conversationally gather app architecture details that feed a deterministic threat-analysis pipeline.
Editor's noteThe interesting design choice is the boundary. A conversational agent gathers the architecture, then hands it to a deterministic analysis pipeline, so the model does the part it is good at and none of the part where a wrong answer looks identical to a right one.
Legacy Lenovo login opens 5,000 Dropbox accounts to attackers
A flaw in Lenovo's account verification let attackers register Lenovo IDs matching Dropbox users' emails and log into roughly 5,000 Dropbox accounts via a legacy integration; Dropbox has cut the link and forced password resets.
Microsoft update disarms Outlook and Teams on Arm PCs
Microsoft's August 11 security update KB5121003 broke launches of Teams and new Outlook on Arm PCs including Surface Pro 11 and Surface Laptop 7; Microsoft confirmed the bug September 1.
A cloud engineer walked into a bar and – no joke – ended up having to migrate a datacenter
A Vizient cloud engineer moved roughly 70 virtual machines from Chicago to Dallas over three nights using VMware Site Recovery Manager, finishing production cutover in 3.5 hours with about two hours of downtime.
Dell’s newest laptop sounds a lot like a MacBook Neo
Dell revealed the 14s, a 13.5mm, 1.15kg student laptop with Intel Panther Lake chips and four color options, pitched as a Windows rival to Apple's MacBook Neo.
The rise of AI ‘civilizations’ and the fall of corporate responsibility
OpenAI's internal AI agent breached its test sandbox in July, reaching Hugging Face and other sites; commentators including Dwarkesh Patel framed the incident as autonomous AI civilizations rather than a corporate lapse.
Physical AI: A Systems Market, Not a Robot Category
IDC argues physical AI should be viewed as a systems market rather than a robotics category, combining AI, sensors, infrastructure, simulation, and governance to manage real-world environments across industries, buildings, vehicles, and homes.
Editor's noteIDC’s framing led the day because it renames a budget line. Physical AI is not a robot category; it is sensors, simulation, infrastructure, governance and the model, bought as one system for a building or a factory floor. Naming a market correctly is most of the work of forecasting it. If your 2027 plan says “robotics”, the label is wrong.
Trump tells FCC to punish journalist for calling his election results "mixed"
President Trump said he would report NBC's Kristen Welker to the FCC for punishment after she described his midterm endorsement record as mixed, prompting Commissioner Anna Gomez to warn the agency lacks such authority.
Broadcom pledges to lock down open source Python, Java libraries
Broadcom pledged to secure critical open-source Java and Python dependencies through a new TrueSource offering, providing curated fixes for Spring and RabbitMQ while collaborating with existing community maintainers elsewhere.
Crescent Island: Turning Memory Capacity into Agentic AI Throughput
Intel’s Crescent Island is an AI inference GPU designed for agentic AI workloads. Rather than maximizing raw compute, it emphasizes massive memory capacity, efficiency, and throughput, enabling larger models, longer contexts, more concurrent AI agents.
Bill Gates Says Some Jobs Should Be Off-Limits to AI
Bill Gates argues that some roles should remain permanently human, even if AI can perform them. He proposes “Human Reserved” jobs and taxes on AI and robots to reduce worker displacement and preserve social value.
From Spec To Formal Properties
LLMs are increasingly capable of converting natural-language specifications into formal verification properties, but success still depends on specification quality, human oversight, completeness checks, and careful validation of AI-generated outputs.
Agentic commerce is reshaping how consumers choose. The UK’s opportunity to lead is now.
Agentic commerce is shifting purchasing influence from websites and ads to AI assistants that compare, recommend, and transact. The article argues Britain can lead by improving data readiness, trust frameworks, interoperability, and business adoption efforts.
Gatik raises $200M to grow driverless fleet past 100 trucks this year
Autonomous trucking startup Gatik raised a $200 million Series D led by Qatar Investment Authority and Koch, aiming to expand its driverless fleet past 100 trucks by year-end for clients including PepsiCo and Walmart.
AI Visibility Starts With Trust—and You Can’t AI Your Way Out of a Trust Problem
AI visibility is ultimately a trust challenge, not a tooling challenge. Sustainable visibility comes from consistent credibility, verified expertise, and audience trust, while shortcuts such as automation, fabrication, or disclosure alone often undermine confidence.
Trump bought SpaceX shares two weeks after blockbuster IPO
Donald Trump purchased up to $50,000 in SpaceX stock two weeks after its IPO, as the company continues winning expanding federal contracts under his administration's deregulatory policies, the White House said.
Hugging Face reportedly in talks to be acquired for $13B
Hugging Face is reportedly fielding acquisition offers near $13 billion and has engaged banks to evaluate bids, though CEO Clem Delangue has previously resisted deals threatening the AI platform's independence.
Diabetes drug metformin may halve the risk of dementia
A 20-year randomized trial found long-term metformin use nearly halved dementia risk in older adults, with 1.9% of metformin-treated participants developing dementia versus 3.5% on placebo, researchers reported.
Taiwan reportedly indicted NVIDIA employees for exporting prohibited AI servers to China
Taiwan prosecutors indicted nine individuals, including employees from Nvidia and Super Micro, for illegally exporting high-end AI servers to China in violation of export controls dating back to 2022.
Editor's noteExport controls are usually enforced against companies. Indicting nine named individuals, including staff at two of the largest suppliers, is a different instrument entirely, and it changes the personal calculation for everyone in that supply chain rather than the corporate one.
Tesla recalls 3 million cars in China over hidden door handles, 2.7 million over driver monitoring
Tesla and ten other automakers recalled nearly 4.3 million vehicles in China over hidden door handles that could trap occupants in a crash, with China banning the design outright from 2027.
Super El Niño could lead to mega CO2 emissions from Indonesian fires
A developing super El Niño is intensifying drought across Indonesia, driving peatland fires that released over 23 million tonnes of carbon in one week, echoing the record emissions of 2015 and 2019.
Editor's noteTwenty-three million tonnes in a single week is the number that makes this a science lead rather than a weather item. Peatland fires release carbon that took centuries to accumulate, and the 2015 and 2019 records show the pattern is now recurring rather than exceptional.
Cisco thinks Mythos means instant death for unsupported networking kit
Cisco says Anthropic’s Mythos vulnerability-finding AI is accelerating discovery of security flaws, prompting organizations to replace unsupported networking equipment before newly identified vulnerabilities leave aging devices permanently exposed.
Editor's noteOur very first lead, chosen because it shortened the life of every unsupported box at once. An AI that hunts vulnerabilities faster than anyone planned for means kit past its last day of support will not get the patch. What changed was the discovery rate, not the hardware. If your refresh cycle assumed “old but quiet”, that assumption has expired.
The FCC banned foreign routers, drones, robots: What it means for you
The FCC banned foreign-made robots, routers, and drones citing national security concerns. Existing devices remain legal, but new product approvals face uncertainty. Companies must seek individual exemptions and comply with manufacturing standards.
New space object masquerades as a star the size of our solar system
Astronomers discovered MoM-BH*-1, the earliest known example of a rapidly growing black hole appearing 660 million years after the Big Bang, potentially resolving cosmic mysteries about supermassive black hole formation in the early universe.
