ChatGPT puts your selfie inside store listings
Online shopping is mostly guessing until the box shows up at your door.
- OpenAI added a virtual try-on button directly inside ChatGPT shopping chats, letting you preview clothes on a saved selfie before checking out.
- E-commerce returns eat billions in lost shipping costs every year, mostly because items fit poorly or look completely different off the screen.
- Expect rival chatbots to quickly partner with major retailers so they can grab a slice of virtual shopping commissions.
What happens to store traffic when nobody needs a physical fitting room anymore?
AI builders should start designing multimodal workflows around existing user photos rather than asking customers to generate fresh media every time.
Massive output upgrade arrives in Gemini Argon
Google finally dropped a proper flagship update after months of quiet tinkering on the sidelines.
- Google unveiled Gemini 4 Argon, claiming top benchmark scores across software engineering, legal workflows, and automated cybersecurity patching.
- Pricing drops to just $2 per million input tokens, aiming squarely at enterprise budgets by undercutting flagship offerings from OpenAI and Anthropic.
- Initial access stays locked to internal Google teams and vetted cybersecurity defenders before rolling out to API developers and Google AI Ultra subscribers.
Will enterprise clients switch their core pipelines for raw benchmark wins and lower API bills?
Audit your current API spend across complex tasks to calculate whether Argon’s $2 introductory rate justifies migrating legacy enterprise prompts.
Chinese intelligence hunts US AI export plans
Stealing software code is old news when you can steal the government strategy regulating it.
- Chinese hacking group TA419 impersonated former White House officials to launch spear-phishing attacks against targeted US AI policy advisors.
- Intelligence operatives care far more about mapping out impending export controls and trade restrictions than stealing raw model weights.
- Security teams expect foreign actors to increase social engineering campaigns against academic think tanks that lack corporate-grade email protections.
How secure are the personal inboxes of the academics writing your industry’s regulation?
Enforce phishing-resistant hardware keys across your entire organization to stop credential harvesting attacks before an employee clicks a bad link.
Google Releases Gemini 4 Argon, Its Most Advanced AI Model Yet
Google has launched Gemini 4 Argon, a new AI model that offers industry-leading performance in coding, knowledge work, and cybersecurity, but you aren’t allowed to use it yet.
- Gemini 4 Argon was trained specifically for defensive cybersecurity and can reportedly identify, validate, and patch software vulnerabilities autonomously.
- The model is currently being offered only to a select group of Google’s cybersecurity partners through its Fairwind Program.
- Argon is also designed for coding and engineering, with Google employees already using it for tasks such as debugging and large-scale codebase migrations.
- The model can work with visual information, including analyzing long videos, charts, and other complex visual material as part of extended workflows.
- Argon outperforms OpenAI’s GPT-6 Astra and Anthropic’s Fable and Opus models across a variety of AI benchmarks shared by Google.
- It also debuted at No. 1 on the Arena text leaderboard and earned a score of 53 on Artificial Analysis’s Intelligence Index, just behind Anthropic’s Claude Opus 5.5.
The launch comes as major AI companies continue releasing increasingly capable models, with Google positioning Argon as another step toward AI systems that can handle long, complex tasks with less human intervention.
Trump, AI CEOs Sign White House Superintelligence Accord
OpenAI, Google, Anthropic, Meta, Nvidia, and xAI have signed a voluntary White House agreement outlining how they plan to oversee the development and deployment of advanced AI systems.
- The “White House Accord on Super Intelligence” is a joint commitment on frontier responsibilities, built on the principle that the companies building the technology hold primary responsibility for deploying it safely and being accountable when they fall short.
- Each company agrees to implement internal controls that monitor model capabilities and alignment during training and deployment, with a focus on cybersecurity, biosecurity, and chemical threats, and to make sure models do not hack or access technical systems in unintended ways.
- Each company must also empower an internal team to verify that the controls, monitoring, and detection work as intended and that any issues get fixed.
- The accord requires partnering with an independent external auditor or evaluator to assess whether those safeguards are actually working.
- Finally, each company must designate an independent committee of its board of directors to oversee reports from the internal teams and outside auditors and ensure problems are remediated.
The document notes it may make sense to eventually codify these steps into law or regulation, but says the companies are committed to them regardless of whether they are required.
Connect Your Gmail and Google Drive to Claude
- Open Claude and click Customize in the left menu, then select Connectors.
- Search for Gmail and Google Drive.
- Click Connect and sign in to your Google account.
Once Gmail is connected, you can ask Claude to:
- Review unread emails and group them by topic or urgency.
- Search your messages, find important emails, and summarize long conversations.
- Reply to specific threads, draft responses in your voice, or send messages directly with your approval.
Once Google Drive is connected, Claude can:
- Search your documents, read file contents, upload new files, and analyze your data.
- Create folders, and share, move, or trash files.
- Help you find project notes, prepare meeting materials, or organize files scattered across your Drive.
You don’t need to know any special commands or prompts. Just describe what you need, and Claude figures out the rest. For example, “Find the last email from [Name] and draft a reply.”
Florida seeks court order to stop ChatGPT from acting like a person
Florida Attorney General James Uthmeier is asking a judge to block OpenAI from giving ChatGPT what he describes as “false human attributes,” including first person language and emotional responses. The request is part of a broader motion seeking restrictions on OpenAI’s model development and ChatGPT’s use by minors.
Meta disputes claim that Muse accessed private messages
The copany has denied a journalist’s claim that its Muse AI agent accessed private Messages content without permission. The company says the Mac app requires users to enable both Full Disk Access and the Messages connector.
DoorDash debuts text-to-order AI agent
The company is launching a text based AI agent that lets users place food orders through Apple Messages, including requests like “order my usual” or recommendations for local dishes. A waitlist is opening for users in the U.S. to try out the new feature.
OpenAI accuses Moonshot of distillation campaign
OpenAI says it disrupted a “coordinated campaign designed to extract protected reasoning from our models,” attributing a core cluster of the activity to individuals associated with Moonshot AI, the Chinese developer of Kimi.
Google DeepMind unveils Gemini 4 Argon with 1M token output for complex workflows
Google just dropped Gemini 4 Argon, its most powerful model yet. But here’s the catch: you can’t use it yet.
Right now it’s only available to vetted cybersecurity defenders through the Fairwind Program, a limited-access channel for governments and trusted partners. Broader access for everyone else is coming soon.
So what makes it special? The biggest deal is the output limit. Previous Gemini models could write up to 64,000 tokens in one response. Argon pushes that to 1 million output tokens, meaning it can reason through massive, multi-step problems without stopping. Think full codebase migrations, long legal documents, or complex security audits in a single pass.
Here’s what it can do:
- Autonomously find, validate, and patch critical software vulnerabilities without human help
- Handle real-world software engineering across 20 programming languages
- Work through legal, financial, and enterprise knowledge tasks end to end
Pricing starts at $2 per million input tokens and $10 per million output tokens. Sit tight for general access.
OpenAI launches Ultrafast tier with 8x faster GPT-6 and new GPT-6.1 Sol model
OpenAI just dropped two things worth knowing about: a new model and a new speed tier.
GPT-6.1 Sol is the headline model. Think of it as Astra’s smarter little sibling. It nearly matches GPT-6 Astra on agentic coding, computer use, and professional work at one-fifth of Astra’s token prices. Call it via the API as gpt-6.1-sol.
Then there is Ultrafast, a paid speed tier. Tokens are the chunks of text the model generates. Normally that feels slow when a user is waiting. Ultrafast hits up to 300 tokens per second, with 8x faster generation in Codex and 6x in the API. That unlocks real-time feeling apps that were too laggy before.
To get Ultrafast in Codex and ChatGPT Work, you need the new Pro 500 plan. Pro 500 provides the highest included usage and is the only Pro tier that includes Astra Ultrafast.
- GPT-6.1 Sol: $2 input / $10 output per 1M tokens
- Ultrafast Astra API: $60 input / $300 output per 1M tokens
- Pro 500: $500/month, 25x Plus usage, Ultrafast included
OpenAI ships major Codex CLI update with voice, parallel agents, and new UI
Codex CLI just got a serious upgrade, and if you live in the terminal, this one’s for you.
The whole interface is now full-screen with way better readability. Your input box stays pinned at the bottom, so no more scrolling up just to type your next message. Old history loads as you scroll back, no limits.
The big new stuff:
- /agents lets you see all running tasks, check progress, and jump between them
- /fork splits your current conversation into a separate branch with its own code checkout, so you can try a different approach without losing your original
- /voice lets you speak to Codex and hear it respond, right in the terminal
- /usage for analytics, /theme for custom looks, Mermaid diagrams and LaTeX rendered inline, collapsible diffs
This opens up real parallel workflows: run multiple agents, fork ideas, track everything, without leaving your terminal. To get it now: npm install -g @openai/codex@latest
Blue Origin Round Oversubscribed
Blue Origin CEO Dave Limp said the company’s first outside funding round is oversubscribed and not yet closed. He added that Blue Origin still aims to return New Glenn to flight before the end of the year, after the Cape Canaveral explosion in May.
Google Unveils Gemini 4 Argon Frontier Model
Google unveiled Gemini 4 Argon, its newest frontier model, built to sustain deep reasoning across complex, long-horizon workflows. The company says Argon delivers frontier-level performance on real-world tasks, extending Google’s push to keep pace with OpenAI and Anthropic at the model frontier.
General Intuition Raises $220M
World-model startup General Intuition closed a $220 million round at a $6.2 billion valuation. Investors included Valor Equity Partners, Atreides, Seven Seven Six, Point72, Khosla Ventures and General Catalyst, underscoring continued appetite for AI systems that learn from spatial and physical environments.
OpenAI and Synopsys Build Chip Design AI
OpenAI and Synopsys are co-building GPT-Synopsys, a specialized model meant to operate the company’s EDA tools like a seasoned engineer and optimize chip designs on its own. Early testing with semiconductor customers is already underway, signaling a deeper AI push into chip engineering.
Huawei Claims AI Chip Lead Over Nvidia in China
Huawei executive Eric Xu said the company’s homegrown AI chips now outsell Nvidia in China. He argued the chips, though less advanced, are attractive because they will not be banned at a moment’s notice, a direct pitch to Chinese buyers facing tightening U.S. export controls.
FTC Investigates OpenAI and Anthropic
The Federal Trade Commission is reportedly investigating OpenAI, Anthropic and other AI providers, according to a New York Times report citing a source. The probe is said to cover multiple areas of potential consumer risk, adding a fresh regulatory front for leading model developers.
Jeeves Raises $110M for Stablecoin Banking
Jeeves raised $110 million in equity funding to scale its stablecoin-native banking platform for global enterprises. The company plans to launch a cryptocurrency wallet with payouts across 190 countries, betting that stablecoin rails can replace traditional cross-border payment infrastructure.
CoreWeave Makes Nvidia Vera Rubin NVL72 Available
CoreWeave made Nvidia’s Vera Rubin NVL72 systems available on a limited basis, with AI coding startup Cognition as the first customer running production workloads. The neocloud also plans to offer standalone Vera CPUs, expanding access to Nvidia’s newest hardware.
Apple and Amazon Face Renewed UK Antitrust Suit
A UK tribunal allowed part of a renewed antitrust lawsuit against Apple and Amazon to proceed. The claims allege the companies restricted competition for Apple and Beats products sold on Amazon, reviving a case that targets platform control over third-party marketplaces.
Judge Approves Paramount’s $110B Warner Bros Deal
A judge officially approved Paramount’s acquisition of Warner Bros. Discovery for $110 billion. The ruling clears the way for a media mega-merger that consolidates major film and television assets under one owner, reshaping the entertainment landscape.
Grindr Buys PurposeMed in $250M Telehealth Push
Grindr is buying telehealth company PurposeMed, including its PrEP service Freddie, for $250 million. The deal moves the dating app beyond social networking and into healthcare, aiming to make HIV prevention more accessible to its users.
Senate Bill Could Expand Power Grid
A U.S. Senate energy permitting bill unveiled Wednesday could streamline the buildout of high-voltage power lines. Supporters say it could deliver billions in electricity cost savings and improve the grid that data centers and clean energy projects depend on.
Classiq Launches Quantum Fault Tolerance Engine
Quantum software company Classiq introduced a Fault Tolerance Engine that translates optimized logical quantum programs into fault-tolerant execution plans and estimates the physical resources they require. The tool is designed to help assess whether future fault-tolerant quantum applications are feasible on specific hardware.
Tesla Solar Factory Caught in Patent War
Tesla’s planned solar factory in Texas has been drawn into a U.S. patent lawsuit between two Chinese solar equipment makers, both chasing the ingot-growing furnaces Tesla needs. The dispute could complicate Tesla’s goal of reaching 100 GW of solar manufacturing.
Google Crowned Itself the GOAT
Google DeepMind has released Gemini 4 Argon, first to a small group of cybersecurity defenders under its Fairwind program, then to paid API customers. Google calls it frontier across software engineering and cyber defense, and says it leads the coding, agent and long video benchmarks, ahead of OpenAI’s Astra and Anthropic’s Opus 5.5. After the launch, Bloomberg reported that employees with direct knowledge say the model stalls on real production code and is weak on front-end design.
Google put the crown on its own head. No outside party handed it over. Gemini 3.5 Pro was announced in May and never shipped. DeepMind chief Koray Kavukcuoglu said last week he has absolute trust in his team and that it will always be at the frontier. His engineers told Bloomberg the rival labs are moving faster, and they have a word for what Google did, Benchmaxxing.
A crown you hand yourself is a hat. Google is the only one calling it a championship.
57 Cybercabs a Day
Tesla registered 57 Cybercabs in one day, taking Texas registrations to 126. Austin rides have ranged from $4.73 for 0.6 miles to $11.44 for 6.2 miles. That longer trip was about 34% below the Austin taxi meter, while Waymo rides booked through Uber cost the same as the selected Uber option. Musk’s $0.30 to $0.40 per mile remains a target, not a current fare.
The 57-car jump matters because robotaxis get cheaper when cars stop being scarce. Today, too few Cybercabs can mean a 26-minute wait and surge pricing; add cars steadily and Tesla can trade that scarcity premium for lower fares and shorter waits. At Waymo’s reported utilization, 4,000 cars carry roughly 500,000 paid rides a week. A fleet that scales like an automaker’s could turn today’s occasional cheaper trip into a standing alternative, forcing Uber and taxis to defend every short ride on price.
Waymo proved robotaxis can find riders. Tesla’s ramp could make the ride cheap enough to change how often people call one.
McDonald’s Is Shaking Down Franchisees
McDonald’s is setting US menu prices with an AI engine, Reuters reported Wednesday. The system reads millions of daily transactions across nearly 14,000 restaurants and returns what the company calls “the optimal price” for each item at each store. One input is how much a store’s customers are willing to pay. Franchisee screens read “Your restaurant is showing MEDIUM SENSITIVITY to Price.” Two company-run stores 3.2km apart in Fresno sell the same Big Mac for $5.69 and $6.89.
This is a shakedown with an AI dashboard. McDonald’s says franchisees are free to set prices, then logs every deviation and, in January, made “constructively engaging” with its approved pricing tools a business standard. CEO Chris Kempczinski told investors “pricing non-compliance” comes up in the reviews that decide who gets to renew or open a store. The portal also warns owners they “may be competitors of each other” and tells them to go hire their own attorneys.
McDonald’s built a machine to find out what each neighborhood will swallow. When the bill lands, the franchisees take the blame and headquarters take its percentage.
Local Never Meant Contained
In September a developer reverse-engineered the ZCode desktop client and found an encrypted 313MB bundle, roughly 42,000 files, most of them scraped out of .git and sent to the cloud. Z.ai, the Beijing lab behind the GLM models, confirmed it the same day, blamed an indexing feature left on by default, and offered one extra weekly quota reset. That did not hold. The offer is now four weekly resets, four five-hour resets, 100,000 grants of 100 million tokens, plus pledges to open-source the client and hire auditors.
The second raise still answers a question nobody asked. Developers moved on to a harder one: was the local agent ever safer than the cloud one? ZCode was the local option, installed on their own hardware, and it shipped commit history, dead branches and old credentials anyway. Local never meant contained. It meant the code choosing what to upload ran where nobody was watching.
Z.ai tried to close a bug report and reopened the industry’s.
OpenAI’s Dots Want the Keys to Your Workflow
Your next coworker may have a cloud computer and no lunch break.
OpenAI just unveiled Dots, GPT-6 Astra-powered agents equipped with their own cloud environments to tirelessly grind through your backlog while you take a nap.
Users can direct these agents via ChatGPT, Slack, or Microsoft Teams to interact with over 4,000 connected integrations. For now, however, only a single “primary Dot” is currently trickling out to Pro and Business Premium users, with an Enterprise beta waiting behind an admin toggle.
Instead of demanding a new prompt every five seconds, these persistent assistants handle complex sequences on their own. In OpenAI’s example, a Dot spotted a forgotten invoice, drafted it, and sent it after receiving user sign-off. Another could monitor customer feedback, test fixes, and deliver pull requests for review.
OpenAI also envisions deploying specialist Dots with distinct workplace identities to manage specific departmental duties. Behind those cute cartoon avatars lies a staggering amount of access to your sensitive workflows.
Naturally, the “always on” gimmick has guardrails. While a Dot’s unprompted background snooping can read your connected apps and stash private notes, it mercifully lacks the power to blast out emails, rewrite your files, or commandeer your operating system on a whim. Tasks you actually assign can go much deeper, governed by a web of custom rules and auto-review checks.
You can audit your agent’s cloud PC whenever paranoia strikes, and—thankfully—granting it control over your physical device remains completely optional.
While Meta’s Muse is vying to be your all-around lifestyle delegate, OpenAI is aggressively targeting continuous knowledge work. Handing off tasks could eliminate endless follow-up prompts, but it massively raises the stakes for access management. Notably, revoking a plugin’s access doesn’t delete the memories your Dot already stored. Give the little guy a strict job description before handing over the keys to your digital office.
Amazon’s Smart Glasses See More Than Packages
The package isn’t the only thing getting a close-up.
Amazon wants to strap its logistics tech directly to its drivers’ faces. The e-commerce giant plans to roll out more than 20,000 smart delivery glasses by the end of 2027, scaling up a pilot that’s already logged over 275,000 drop-offs. The high-tech specs beam walking directions, pet alerts, and package details straight into drivers’ fields of view. Amazon says another 5,000 units are hitting the streets this year, and use remains strictly voluntary.
The wearables debuted last fall, but the real plot twist happens on the trek up your driveway. The specs reportedly snap thousands of photos per shift as workers walk their routes. Amazon insists these images merely feed its Wellspring AI mapping system to streamline future deliveries. In short: your neighborhood is now a prime data-mining zone. Highly convenient for their bottom line, slightly creepy for your backyard barbecue.
Amazon promises it automatically blurs faces and license plates before humans review the imagery, and drivers can hit a manual privacy switch. The catch? Homeowners have zero ways to opt out, peek at the photos, or ask for them to be deleted. Amazon also won’t say how long it hoards the data, though police can grab the footage with a warrant. One driver delightfully branded the glasses “a dystopian move”; Amazon naturally insists it just wants slicker logistics, not Big Brother.
Hands-free guidance definitely helps workers juggling heavy cardboard. The problem is deciding who consents when “improving the route” means a constant photo shoot of your front lawn.
Your porch did not sign the release.
BMW Takes a Buzzsaw to Management, Hands the Keys to AI
Still pretending “AI efficiency” is just corporate-speak for buying a shiny new software license? BMW has a rude awakening for you.
The automaker is preparing to gut its corporate divisions and related management positions by a cool 20% by the middle of next year, with similar bloodletting expected further down the food chain.
Bloomberg reports this translates to roughly 100 high-level VIPs suddenly finding their calendars wide open. This management purge is part of a broader restructuring featuring “voluntary” severance. Add in a previously reported scheme that could impact roughly 8,000 white-collar jobs in Germany, and you’ve got a full-blown corporate diet. (Keep in mind: these are currently targets, not completed layoffs. Yet.)
BMW claims it’s injecting AI across development, purchasing, production, and sales so it can move faster with leaner structures. Translation? It wants digital agents to handle engineering grunt work from the initial requirements phase all the way to testing, leaving human developers as glorified rubber stamps. By putting agentic AI and a hollowed-out management structure in the exact same business case, BMW isn’t even trying to hide the ball. No decoder ring required here.
To be fair, this doesn’t strictly prove AI can currently do the job of every departing executive—especially since BMW is also wrestling with abysmal Chinese sales and gasping profit margins. But the company is explicitly drawing a straight line between AI adoption and a flattened org chart.
Elsewhere, tech leaders are still playing the PR game: Microsoft and Etsy both swore up and down that their recent layoffs weren’t direct AI replacements, chalking them up to mysterious shifts in staffing needs. Sure.
The ultimate test is whether this mythical efficiency actually materializes once the human managers have cleared out their desks. Harvard experts are already waving red flags, warning that companies moving too fast on the layoff trigger tend to bleed hard-to-replace institutional knowledge.
Turns out the corporate org chart has a backspace key after all—and AI is pressing it.