AI and ML

Standalone experiment killed after less than 12 months as model maker redirects agentic ambitions towards workplace productivity

OpenAI has decided its AI browser experiment has run its course, pulling the plug on ChatGPT Atlas less than a year after launch and moving its browser-based agent features into ChatGPT and Codex.

The company said Atlas will stop working on August 9 as it rolls out the newly unveiled ChatGPT Work platform.

Atlas arrived last October with no shortage of ambition. Rather than trying to out-Chrome Chrome, OpenAI wanted to bolt ChatGPT directly onto the web, promising a browser that could read pages, rewrite them, answer questions, and eventually start doing the clicking itself.

However, within days of its debut, security researchers demonstrated prompt injection attacks that could manipulate the browser’s AI assistant into following malicious instructions embedded in web pages. A few days later, researchers uncovered another flaw that allowed malformed URLs to cause Atlas to expose information about previously visited sites. Neither flaw proved fatal, but they quickly exposed the gap between an AI browser on launch day and one ready for the open web.

OpenAI hasn’t fully abandoned the idea of AI-powered browsing, but appears to have decided it doesn’t need a standalone browser to deliver it.

ChatGPT Work looks less like a browser and more like OpenAI’s attempt to become the operating system for office work. The desktop application combines ChatGPT, Codex, and Atlas into a single package that can connect to files and business applications, browse the web, generate documents, spreadsheets, presentations, and websites, and stick with long-running projects for hours at a time instead of answering one prompt after another.

Powering it all is GPT-5.6, OpenAI’s latest model series, which the company says is better at reasoning through multi-step tasks and producing work that follows users’ templates and reference material.

The pitch is different this time around too. Atlas was all about rethinking the browser, but ChatGPT Work is about getting office workers to spend more time with OpenAI. To help with that, OpenAI has bundled plugins into a single directory, allowing ChatGPT to pull context from tools including Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRMs, and project trackers, either automatically or when users explicitly call them into a prompt. 

For developers, the biggest change is that Codex is no longer treated as a separate product. Codex is losing its standalone app and moving into ChatGPT Work, picking up inline diff editing, pull request reviews, and multi-repository support.

ChatGPT Work is available to all plans on desktop, rolling out to Plus, Pro, Business, Enterprise, and Edu on web and mobile over the next few days.

AI browsers have attracted no shortage of hype over the past year, but convincing people to swap Chrome for an AI-first alternative was always going to be a taller order than bolting another chatbot onto the web.  ®

Looking for the most recent Connections answers? Click here for today’s Connections hints, as well as our daily answers and hints for The New York Times Mini Crossword, Wordle, Connections: Sports Edition and Strands puzzles.

Today’s NYT Connections puzzle was a true mix. The blue category words jumped right out at me and the purple category is really creative when you realize what it is. Read on for clues and today’s Connections answers.

The Times has a Connections Bot, like the one for Wordle. Go there after you play to receive a numeric score and to have the program analyze your answers. Players who are registered with the Times Games section can now nerd out by following their progress, including the number of puzzles completed, win rate, number of times they nabbed a perfect score and their win streak.

Read more: Hints, Tips and Strategies to Help You Win at NYT Connections Every Time

Hints for today’s Connections groups

Here are four hints for the groupings in today’s Connections puzzle, ranked from the easiest yellow group to the tough (and sometimes bizarre) purple group.

Yellow group hint: The big top.

Green group hint: Like a placid lake.

Blue group hint: To infinity, and beyond!

Purple group hint: Repeated letters in special spots.

Answers for today’s Connections groups

Yellow group: Circus equipment.

Green group: Undisturbed, as water.

Blue group: Toy Story characters.

Purple group: Double letters appearing in that letter’s alphabetical position.

Read more: Wordle Cheat Sheet: Here Are the Most Popular Letters Used in English Words

What are today’s Connections answers?

The yellow words in today’s Connections

The theme is circus equipment. The four answers are cannon, stilts, trapeze and unicycle.

The green words in today’s Connections

The theme is undisturbed, as water. The four answers are calm, flat, glassy and still.

The blue words in today’s Connections

The theme is Toy Story characters. The four answers are Bo Peep, Jessie, Slinky and Woody.

The purple words in today’s Connections

The theme is double letters appearing in that letter’s alphabetical position. The four answers are aardvark, bocce, ebbing and twiddle.

U.S. federal cybersecurity agency CISA said it did not have a prepared response plan for how it should handle a cybersecurity incident in May, after an investigative reporter notified the agency that a contractor had publicly exposed sensitive keys and credentials for accessing U.S. government systems.

CISA, the Homeland Security unit tasked with defending federal networks and helping to safeguard critical infrastructure, revealed Friday in a postmortem report that its staff “had to spend time building [a playbook] during the early stages of the incident.” The agency said it is important to prepare playbooks for “all anticipated needs” to ensure that organizations are ready to respond in the event of a security incident rather than scrambling to improvise one in real time.

The agency did not say how long the missing playbook delayed CISA’s response, and a spokesperson did not immediately respond to TechCrunch’s request for comment. 

Independent cybersecurity journalist Brian Krebs reported in May that a security researcher with cyber firm GitGuardian alerted him to reams of exposed passwords stored in a publicly accessible GitHub repository, which an employee of a CISA contractor had uploaded.

According to Krebs, the researcher tried to alert the contractor but didn’t hear back. Only after Krebs contacted CISA did the agency take the repository offline and revoke and replace all of the exposed credentials to prevent any potential future abuse.

CISA said that no customer or mission data was exposed in the incident and thanked the researcher and reporter for their help. The agency said that its channels for allowing security researchers to notify CISA of potential incidents “were not well defined,” and that it has made changes to make it easier and faster for researchers to contact the agency.

CISA has been without a permanent director since the start of President Donald Trump’s second term in January 2025. The agency has also been affected by cuts, furloughs, and layoffs affecting about a third of its workforce since Trump took office. 

The Dutch National Police (Politie) says it has found “strong indications” that Dutch hackers have been involved in a February breach at the telecommunications provider Odido.

“This includes a telephone conversation that was made with Odido customer service shortly before the hack. In this conversation, a Dutch-speaking man posed as Odido’s IT employee. The company was then misled through phishing, after which the data theft took place,” the police said in a Thursday press release.

“This type of investigation is often complex and takes time, but cybercriminals are also vulnerable and leave traces. Traces have been secured at several times during the investigation into the hack at Odido, which the research team continued to work on,” added Stan Duijf, the head of operations at the National Investigation and Interventions Unit.

Odido is one of the largest Dutch telecommunications companies, offering mobile, broadband, and television services to millions of customers across the Netherlands.

When it disclosed the breach on February 12, the company said the attackers accessed its customer contact system on February 7 and downloaded the personal data of many of its users. It also told local media that the resulting data breach affected 6.2 million customers and that the threat actors reached out to say they had stolen millions of user records.

According to the telecom firm, the exposed information varies per customer, and it may include a combination of full name, address and city of residence, mobile number, customer number, email address, IBAN (bank account number), date of birth, and some identification details (passport or driver’s license number and validity).

However, it added that no call details, location, data, billing data, scans of identity documents, or Mijn Odido passwords were exposed during the incident.

While Odido has yet to attribute the incident, the ShinyHunters extortion gang claimed responsibility for the breach on its dark web leak site, releasing an 88GB archive containing over 15 million records, including data the company had already disclosed as exposed in the attack.

​ShinyHunters has been behind widespread vishing campaigns targeting Okta, Microsoft, and Google single sign-on (SSO) accounts, impersonating IT support staff to trick targets’ employees into entering credentials and multi-factor authentication (MFA) codes on phishing sites.

After breaching corporate SSO accounts, the threat actors steal data from connected SaaS applications, including Microsoft 365, Google Workspace, Salesforce, SAP, Slack, Zendesk, Dropbox, Adobe, Atlassian, and others.

The cybercrime group has been linked to a growing number of breaches involving companies such as Google, Cisco, PornHub, the online dating giant Match Group, the European Commission, Rockstar Games, and the McGraw-Hill edtech giant.

They were also behind security breaches at over a dozen Snowflake customers, various other third-party integration providers, and, more recently, a new series of breaches that hit over 100 organizations (including the University of Nottingham) following data-theft attacks exploiting an Oracle PeopleSoft zero-day flaw.

Security teams log 54% of successful attacks and alert on just 14%. The rest move through your environment unseen.

The Picus whitepaper shows how breach and attack simulation tests your SIEM and EDR rules so threats stop slipping by detection.

Get the whitepaper

Bigme built the HiBreak Dual 2 around a simple mechanical solution to a familiar problem. Regular smartphones deliver color, speed, and apps but wear out eyes during long reading sessions. Dedicated e-readers protect vision and sip power yet lack cameras, video, and quick multitasking. This device carries both screen types in one slab so owners turn it over instead of juggling separate gadgets.

The 6.13-inch E-ink display is available on one side. The black and white version measures 824 by 1648 pixels, which is a fairly acceptable resolution of roughly 300 pixels per inch. Color versions aren’t as sharp, but they do offer thousands of distinct colors without the need for a larger screen. Bigme rates the screen for up to 80 hertz refresh in supported modes. The front light has 36 brightness settings and can be tuned to warm or cool, making reading for extended durations more pleasant whether indoors or outside in bright sunlight. You can even write and draw straight on the screen; simply grab an optional pressure-sensitive stylus and you’re ready to go.

Sale

XTEINK X3 eBook Reader (Developer Edition), 3.7″ Pocket E-Ink eReader, Ultra-Thin 0.2″, 2 oz…

• 3.7″ Pocket eBook Reader, Only Approx. 58g: Take your library anywhere with XTEINK X3, a compact 3.7-inch pocket eReader designed for true everyday…
• Paper-Like E-Ink Reading, Made for Focus: Enjoy a clean, paper-like E-Ink reading experience that feels gentle on the eyes and helps you stay focused…
• Gyroscope Page-Turn + Physical Buttons: Read comfortably with one hand using gyroscope page-turn control and responsive physical buttons. Whether you…

Flip it over and on the other side, you’ll find a 5-inch LCD. That’s 1280 by 720 pixels of full-color action, ideal for viewing videos, scrolling through social media, playing games, or even using the camera, since the 50-megapixel sensor on the back works well with this display for framing images. You can also use your face to unlock the phone.

The MediaTek Dimensity 8300 processor handles everything on both screens. You can choose 12 or 16 gigabytes of RAM combined with 256 or 512 gigabytes of storage. Android 16 and Google Play are pre-installed, so you have the complete package. There are unique software fixes available to help apps work properly when you flip the smartphone around. Dual SIM 5G, NFC, Bluetooth 5.3, and stereo speakers complete the connectivity possibilities. The battery has 4450 milliamp hours and a 30w quick charger, albeit the E-Ink side does not draw much power as you sit there reading for hours on end. That will help you get more use out of your battery. Of course, if you use the LCD side frequently, your battery life will suffer, but Bigme believes you should be able to get through the day just fine.

Bigme improved on their previous dual-screen model. They ended up with a larger LCD around the back and a greater refresh rate on the front, but the essential idea remains the same. It’s still a physical flip, not simply software trickery. They also solicited user suggestions on how to improve the software’s usability, and it appears like they did a decent job of keeping the various aspects of the phone separate.

On Kickstarter, the base black and white model costs $599 and includes 12 gigabytes of RAM. If you want a color version or 16 gigabytes of RAM, you’ll need to pay $200 more. Each pledge includes a protective cover, and you can purchase the stylus separately. Bigme expects to begin distributing these out this fall, before the rest of the globe can get their hands on them at exorbitant retail pricing.

An enterprise AI agent answers with total confidence, but the number is wrong. Nobody catches it until someone traces it back to a stale metric definition or a document the retrieval system never pulled. The model did not fail. The context it was given did.

In the past six months, 57% of enterprises traced a confident but wrong AI agent answer to missing or inconsistent business context, and 31% said it happened more than once, according to a VB Pulse June 2026 survey of 101 qualified enterprises with more than 100 employees.

The reason is not hard to find. Retrieval over documents is the default way agents get business context for 38% of enterprises, nearly double the next closest approach. The way most enterprises choose a retrieval system compounds the problem. Ease of ingestion and operational simplicity lead the selection criteria, with retrieval accuracy running behind both. The accuracy problem only shows up after the system is already live.

There is a known fix for this, a governed context layer every agent reads from instead of guessing. Vendors are racing to roll out context platforms while most enterprises are still figuring out what it is.

75% don’t have an agentic context layer yet

The context layer is meant to be a shared model of what business data actually means, built once and referenced consistently instead of re-derived by every agent that touches it. 

The VentureBeat research shows the enterprise response to that idea is broad but unfinished. Twenty-five percent of respondents run one in production. Thirty-four percent are building one right now. The remaining 41% have not started.

Among companies already building or running a governed context layer, 78% report a confident-wrong failure — an AI agent that answered with total certainty and was still wrong. Among companies with no plans to build a layer, only 20% report the same thing. Companies that already got burned are far more likely to be building the fix. Companies that haven’t been burned yet see no urgency.

What governed context looks like when someone actually builds one

Every major data and AI platform vendor is now building some version of this layer, and they are not converging on the same architecture. 

• DataHub is treating catalog metadata and years of analyst query behavior as a knowledge source, then keeping it current as a living system rather than a static wiki. 

• Microsoft’s Fabric IQ is building a business ontology that any agent, not just Microsoft’s own, can query over MCP. 

• Couchbase is pushing agent memory and context retrieval down to the edge, arguing the operational database is a more natural home for it than a search or analytics layer bolted on after the fact. 

• Pinecone’s Nexus is compiling structural logic into the metadata layer ahead of runtime, betting that agents need pre-built structure more than they need faster search.

• Snowflake runs a two-layer system, Horizon Context for customer-managed definitions and Cortex Sense for context the platform infers on its own. 

• Oracle’s Unified Memory Core takes the opposite approach, folding vector, graph and relational data into one transactional engine so there is no sync layer left to go stale. 

• Google’s Knowledge Catalog mines query logs and usage patterns to curate semantic context automatically.

• AWS’s Context service makes the same bet, a knowledge graph that gets smarter from how agents actually use it rather than from manual re-curation.

Analysts converge on one diagnosis

The vendor approaches differ. What analysts and practitioners have told VentureBeat about the underlying problem, across a run of interviews this year, does not.

When DataHub’s context layer push landed this spring, Constellation Research VP and principal analyst Michael Ni framed the stakes in blunt terms. “Whoever controls runtime context controls the AI decision layer for enterprise data,” Ni said. He was equally direct about how far any single product actually gets a buyer. “Vector memory isn’t business meaning, business meaning isn’t governance and governance isn’t execution,” Ni said.

In the same interview, BARC analyst Kevin Petrie pointed to a narrower but concrete gap. Most context platforms concentrate on structured tables, he said, which give agents trusted facts but miss the harder, messier context locked in documents and unstructured content, exactly the material a business actually runs on day to day.

Stephanie Walter, practice leader for AI Stack at HyperFRAME Research, made a related point earlier this year when VentureBeat asked her about enterprise context fragmentation. 

“The market is converging on the same conclusion,” Walter said. “Agents don’t just need more tokens or better models. They need governed, current, low-latency context.” She made a similar case in an earlier review of Pinecone’s Nexus launch, careful not to overstate how new any of this is. Nexus, she said, “shifts knowledge work from runtime chaos to pre-compiled structure. But it’s an evolution of RAG architecture, not a complete reinvention.” 

Gartner’s Arun Chandrasekaran, reviewing the same launch, offered the more forward-looking read. Agentic AI, he said, is moving from pure information retrieval toward a reasoning architecture, one where long context works as short-term memory and a vector database functions as deep storage underneath it.

The fragmentation problem shows up hardest at the practitioner level, where separate tools for retrieval, memory and access control were never built to agree with each other. Steven Dickens, CEO and principal analyst at HyperFRAME Research, put it bluntly after Oracle’s AI database push landed this spring. “Data teams are exhausted by fragmentation fatigue,” Dickens said. “Managing a separate vector store, graph database and relational system just to power one agent is a DevOps nightmare.” 

Matt Kimball at Moor Insights and Strategy, in that same story, put the production reality more simply. Getting an agent working is not the hard part, he said. The struggle is running it in production, where the goal becomes removing the distance between data and execution rather than adding another layer on top of it.

What this means for enterprises

Here’s what this adds up to for enterprises building on this layer.

Retrieval alone will not close the context gap. RAG is the default source for context in most enterprises today, and it is also the layer most closely associated with the confident-wrong-answer failure. Adding more documents or a bigger index does not fix a definition that is inconsistent across systems.

The semantic context layer is where the budget is actually moving, even where it hasn’t shipped. Fifty-eight percent of enterprises are already engaged — building or in production — but only 25% have actually gotten a layer live. That gap shows where enterprises have decided to spend, not where they’ve arrived.

No single vendor owns the architecture yet, and that is likely to stay true for a while. Enterprises evaluating this layer should expect to integrate rather than pick a single winner, at least for the next several quarters.

The buying decision is happening this year, and it is concentrated among the companies already burned by it. Fifty-seven percent of enterprises plan to switch or add a retrieval or context platform within the next twelve months. That intent is not spread evenly. Enterprises that reported a repeat confident-wrong failure plan to switch or add a provider at roughly 81%, against 32% among enterprises that never hit the problem. The companies shopping for new context tooling right now are largely the ones whose agents already got it wrong.

The agents are already running. The context underneath most of them is still being built, and the vendor selling the fix is being chosen this year.

This data will be part of a broader conversation at VB Transform 2026 on July 14 and 15 in Menlo Park: the context gap enterprises are racing to close, and which of the emerging approaches — governed semantic layers, hybrid retrieval, provider-native bundles — actually holds up in production.

Looking for the most recent Wordle answer? Click here for today’s Wordle hints, as well as our daily answers and hints for The New York Times Mini Crossword, Connections, Connections: Sports Edition and Strands puzzles.

Today’s Wordle puzzle is tricky, with one very unusual letter included. If you need a new starter word, check out our list of which letters show up the most in English words. If you need hints and the answer, read on.

Read more: New Study Reveals Wordle’s Top 10 Toughest Words of 2025

Today’s Wordle hints

Before we show you today’s Wordle answer, we’ll give you some hints. If you don’t want a spoiler, look away now.

Wordle hint No. 1: Repeats

Today’s Wordle answer has one repeated letter.

Wordle hint No. 2: Vowels

Today’s Wordle answer has two vowels, but one is the repeated letter, so you’ll see that letter twice.

Wordle hint No. 3: First letter

Today’s Wordle answer begins with A.

Wordle hint No. 4: Last letter

Today’s Wordle answer ends with N.

Wordle hint No. 5: Meaning

Today’s Wordle answer can refer to anything that relates to, resembles or is derived from birds.

TODAY’S WORDLE ANSWER

Today’s Wordle answer is AVIAN.

Yesterday’s Wordle answer

Yesterday’s Wordle answer, July 10, No. 1847, was CANAL.

Recent Wordle answers

July 6, No. 1843: TODDY

July 7, No. 1844: SLING

July 8, No. 1845: DEMON

July 9, No. 1846: AMEND

Engineers at 1X put the motors for the new NEO hands up in the forearm. Strong cables run down through the wrist and pull on the fingers and palm. This tendon-driven layout keeps the hand itself light while delivering strong, precise pulls. Low gear ratios between 5-to-1 and 15-to-1 let the system stay backdrivable. Push on any finger and it yields while reporting the exact force it encountered.

This design tweak allows the hand to engage with the environment in a much more natural manner. Robot hands typically use high gear ratios, which make them stiff and entirely numb to any contact. The sensation of force is lost within the machinery before it can be felt. Here, the relationship is direct. Every joint acts as an actuator/sensor combination. This allows the hand to probe and learn from each touch without the need for additional gear on top. The fingertips are outfitted with some very high-resolution tactile sensors that read normal force, contact location, and even shear pressure. That shear sensing feature is fantastic; it detects the first warning that something is ready to slip. We’ve witnessed the hand detect a glass beginning to slide and tighten up before it just barely falls. The same sensors also provide some extremely fascinating visualizations that illustrate pressure maps and where touch is during a greeting or while attempting to pick up some fragile origami.

Unitree R1 Humanoid Robot (White, R1 Air)

• Three models, one lightweight platform R1 Air (20 DOF, monocular camera), R1 (26 DOF, binocular camera, head+waist joints), and R1 Edu (26 DOF…
• Easy setup – no coding required for basic use Unbox, power on, and start. Manual teaching feature: physically pose the robot, and it replays the…
• More DOF = more expressive movement 26‑DOF models (R1 / R1 Edu) add head and waist articulation for smoother dance and running. For safety reasons…

Speed is especially noticeable in live demos, as fingers drum against a surface so quickly that you can barely see what’s going on. People in the audience yell out faster and faster rhythms; the hand maintains a steady pace before slamming on the brakes as instructed. That rapid responsiveness combined with a very wide range of motion is an excellent combination. When the occasion demands for it, fingers can stretch far beyond what a human hand can do while remaining close enough to human movement to allow operators to operate with the hand safely without colliding.

Real-world tasks, on the other hand, truly demonstrate the capabilities of this upgrade. It is perfectly capable of picking up a single screw or coin from a flat surface. It can sort grapes by color without damaging them. It can turn on a light bulb, use a screwdriver, zip up a jacket, connect a USB-C connection, pour tea from the kettle, and even make a respectable sign in American Sign Language. Delicate and powerful elements coexist on the same platform, making this a true all-purpose hand. Peak torque is 3.5 Nm at the base of the thumb and 2.6 Nm at the primary finger joints, with distal flexion forces of up to 45 N. The wrist adds 17.75 Nm, and positioning precision is ±0.2 mm.

Then there’s the durability testing, which showed that entire finger assemblies and drive units could withstand millions of cycles without failure. The wrist joints alone performed nearly 2 million cycles under strain. So, the entire hand is completely waterproof, rated IP68, and made of food-safe materials. Another advantage of NEO is that it can just rinse its own hands under the faucet, just like you would.
[Source]