Ultra-level features and a few design tweaks highlight the updates.

What’s new on the second-generation QuietComfort Headphones?

Billy Steele for Engadget

What’s good about these headphones?

Billy Steele for Engadget

What’s not so good about the new QuietComfort Headphones?

Billy Steele for Engadget

Wrap-up

AI is already changing how people find and buy things online, comparing prices, narrowing down options, doing in seconds what used to take an afternoon of tabs open.

An AI agent could find a flight that fits your schedule, for example. Giving it permission to actually buy the ticket is a more complicated problem.

That is because most payment systems were built around people. Someone enters their card details, signs into an account, clicks approve. If software is going to complete more tasks on their own, they will also need a way to pay without stopping to ask permission every time.

There are signs this could become a much bigger part of how software works. Gartner expects agentic AI to be built into 33% of enterprise software applications by 2028, up from less than 1% in 2024. McKinsey estimates that AI agents could mediate between $3 trillion and $5 trillion in global consumer commerce by 2030.

How those agents actually pay is now becoming its own technology problem. Mastercard, Coinbase, and blockchain networks, including XDC, are among those trying to solve it.

When software needs to buy something

Mastercard is preparing for the same possibility. When it launched Agent Pay for Machines in June, the company described a future in which agents could continuously buy services from each other, including payments worth fractions of a cent.

That is where XDC AI also comes in: by providing AI agents with the capability to find and pay for digital services on demand, rather relying on traditional human checkout flows, the platform allows software to programmatically handle micro-transactions in real time.

“Every internet transaction so far has assumed a person is on the other end of it,” said Atul Khekade, co-founder of XDC Network. “AI Agents are breaking that assumption completely. If software is going to act on our behalf, it needs a way to pay for what it needs without waiting for someone to approve it every time.”

A user can give an agent access to a wallet and set a limit on how much it is allowed to spend. When the agent needs data from a paid API to finish the task, it can make the request and pay for it in USDC, with XDC covering the gas fee, so the agent does not also need to hold XDC to make the payment.

XDC Tech, the US institutional arm of XDC Network, has also integrated with Bridge, a Stripe company, helping to connect those onchain payments with more traditional financial infrastructure by giving developers access to infrastructure for moving between fiat currencies and stablecoins.

An agent researching a market could need information from several paid data sources. Another might need access to a particular API or a small amount of computing power to finish a task. Instead of someone setting up a new account or subscription every time, the agent could find the service and pay only for what it needs.

An old part of the web finds a new use

XDC is building this on x402, an open protocol developed by Coinbase. That resurrects HTTP 402, an old web status code meaning “Payment Required” that has existed for years without becoming widely used. An agent requests a paid API, the server responds with a price instead of the data, and the agent’s wallet pays automatically, no login, no card number, no human in the loop. XDC has built that flow into its own marketplace, where an agent can find an API, pay per use in USDC, and stay within whatever limit its owner set.

Adoption is still early, but x402 is beginning to attract some large names. The Linux Foundation now oversees the x402 Foundation, which has 40 members including Mastercard, Visa, Stripe, American Express, Google, AWS and Circle. That does not mean x402 has become a mainstream payment rail, but it gives the protocol considerably more institutional backing than it had when Coinbase first introduced it.

That is a major reason why the infrastructure work matters more than the headlines right now. Fully autonomous shopping is still limited, and AI is much further along at helping people compare options than at completing purchases on its own. But the direction is clear, and XDC is betting that when agents do start paying their own way, they will need rails built for machine speed rather than human checkout, cents instead of dollars, and thousands of transactions instead of one.

The gap between x402’s onchain activity and its ecosystem valuation is a reminder that infrastructure and adoption rarely move at the same pace. What is being built is real, the settlement rails, the wallet permissions, and integrations with players like Bridge, but the volume moving through them today is still a fraction of what the numbers around this space suggest.

For now, AI remains far better at helping people decide what to buy than at completing the purchase itself. That is where the real contest is playing out. Card networks are wagering that trust and identity, the problem they have spent decades solving for humans, will matter just as much for machines.

Networks built for stablecoin settlement, XDC among them, are wagering the opposite: that speed and near-zero fees will define this market once agents start transacting at scale. In the end, the better approach will probably come down to the use case. But what is certain is that the infrastructure decisions being made now, by companies like Mastercard and XDC, will shape how autonomous commerce actually works once it arrives.

Andela collected data from 47,101 technical job postings from Fortune 500 companies and 2,026 skill descriptions. 

New research from global talent platform Andela has found that in a working landscape being transformed daily by AI, organisations are failing to update their job descriptions and titles in line with changes. 

Andela’s ‘Emerging Skills Research’ collected data from 47,101 technical job postings from Fortune 500 companies and 2,026 skill descriptions. 

A core finding of the research was the revelation that 53pc of AI job postings are asking for skills that don’t match the listed job title, highlighting a disconnect between organisations, recruiters and jobseekers. 

“AI is changing tech workflows faster than ever and we need new ways to get ahead of those changes,” said Carrol Chang, Andela’s CEO. 

“This groundbreaking research, based on our proprietary skills taxonomy updated for the AI world, provides a headlight, not a rear-view mirror.

“With it, companies will identify emerging roles and skills ahead of the market, enabling both enterprises and employees to be more strategic and successful.”

Rambling roles

Andela’s research found that among the roughly 1,832 postings targeting people with titles such as AI engineer and ML engineer, more than half of the required skills pulled from at least two different and established roles. 

The report said: “For familiar titles like AI engineer, companies are seeking a different, unnamed role that includes skills such as LLM orchestration, autonomous agent and vector databases. Or, companies hire data scientists but then ask them to ship LLM agents.”

Reviewing job listings from Fortune 500 companies, Andela found that 23 ‘groups’ of skills kept appearing together, but without any official job title being attached. Eight of the groups appeared to suggest a brand new kind of job, 14 were found to be hybrids of old jobs with new skill expectations, and one was dropped due to a lack of confirmed data. 

Two examples of ‘new jobs’ discovered are MLOps pipeline engineer, which combines parts of five existing job roles into one, and LLM application engineer, where someone builds apps using AI models that already exist, instead of building those models from scratch.

Andela found 6,758 job listings essentially asking for an LLM application engineer – they listed all the relevant skills, but the companies didn’t use that job title because it doesn’t officially exist yet. This shows companies need these skills faster than the job market has caught up with proper names for them, according to the research.  

The report said: “The findings underscore the challenges companies face, driven by the unprecedented speed of change in the post-AI era, to accurately identify emerging skills, name them and attach them to job roles. Mismatches result in lost time, money and opportunity for both employer and employee. 

“Job descriptions written for yesterday’s roles filter out the candidates companies actually need. Companies that fail to see emerging needs for human skills, especially as AI changes what it can do, will continually chase the wrong thing, and companies who hire today for yesterday’s roles will likely accrue ‘talent debt’ alongside technical debt.”

According to Cory Hymel, an author on the report and Andela’s head of research, gaps that widen the divide between what companies think they are hiring for and what the role truly requires are becoming structural liabilities. 

As of now, there are a number of emerging tech roles as identified in the research, which include: MLOps pipeline engineer; LLM application engineer; FinOps reliability engineer; docs-as-code engineer; product front-end engineer; lakehouse analytics engineer, DevSecOps security engineer; and SecOps observability engineer.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

• The Badkruka is IKEA’s newest affordable Bluetooth speaker
• It has Spotify Tap integration and physical controls
• You can pair multiple speakers together in a variety of setups

It seems IKEA is on a roll when it comes to cube-shaped, low-price Bluetooth speakers: following on from the Ikea Kallsup that we tested earlier this year, we now have the IKEA Badkruka, which is several times bigger and offers some extra features.

There’s no official announcement about the Badkruka just yet, so we’re relying on the product page on IKEA’s German website (via Notebookcheck) for all the details. It also shows up on the Swedish site, but not the US, UK, or Australia yet.

SYSTEMS

Shipments rise with enterprise and government buyers joining the hyperscaler spending spree

While high memory costs have hurt PC shipments, the server market continues to grow as AI infrastructure spending spreads beyond hyperscalers to corporate and government buyers.

According to market intelligence firm IDC, the second quarter was a bumper one for the server sector, with vendor revenue reaching an all-time high of $166.3 billion. That was a 52 percent increase from the same period last year.

The picture for servers therefore differs from that for laptops and desktops. There, unit shipments have fallen as buyers are discouraged by higher prices, driven by shortages of memory components. Yet higher prices have helped larger vendors sustain their revenue.

In contrast, server shipments increased by 15.4 percent year-on-year in Q2, despite average selling prices being pushed up by elevated memory pricing and continued supply issues with other components. IDC said average selling prices increased across both GPU-accelerated and non-accelerated systems.

Average selling prices for GPU-accelerated servers rose by nearly 44 percent to $170,200, even as GPU unit shipments fell 10.8 percent year-on-year. For non-accelerated systems, average pricing was up by more than 33 percent to nearly $13,000.

AI infrastructure investment from hyperscalers and large cloud providers remains the largest source of demand, IDC observes. GPU-accelerated servers for the AI market made up nearly 53 percent of total revenue during Q2.

However, it also says that AI server adoption is broadening beyond the largest players into enterprise and government-directed deployments across a growing number of countries, a policy and capex-driven layer of demand that is largely insulated from near-term commercial budget cycles.

“The notable shift in the server market this quarter is in who is now buying,” said Kuba Stolarski, IDC research vice president for Computing Platforms and Service Provider Infrastructure.

“Demand is broadening beyond the largest hyperscalers toward specialized cloud providers (or neoclouds), sovereign AI programs backed by public capital, and enterprises beginning to adopt agentic and inferencing workloads,” he added.

Non-x86 servers now account for 44.8 percent of all server market revenue, according to IDC. That share has fallen from the first quarter, when they made up nearly half the total, despite the actual revenue figure rising from $58.7 billion to $74.4 billion.

Another trend highlighted by IDC is that the big brands are starting to eat into the share of original design manufacturers (ODMs), the so-called white box server makers that have traditionally met the requirements of the hyperscalers.

While ODMs collectively still make up the lion’s share of server market revenue, this fell from over 60 percent last year to 53.9 percent in Q2. Leading the way is Dell Technologies, whose share rose from 7.7 percent a year ago to 13.4 percent. Supermicro is the second largest player, with 6.1 percent, followed by Lenovo on 5.1 percent, while HPE came fourth with 3.5 percent.

The United States remains the biggest server market, generating $112.2 billion in Q2, or 67.4 percent of global revenue. China generated $26.4 billion, while Asia-Pacific excluding China and Japan reached $10.9 billion. Western Europe generated $9.1 billion and Central and Eastern Europe $0.7 billion. ®

No matter how you feel about the iPhone Duo and its exorbitant $1,999 price, its existence proves that Apple has finally figured out how to make a foldable. And on a broader level, it also shows that Apple isn’t too proud to follow in the footsteps of Samsung and other Android phone makers. If that’s the case, I’m hoping Apple also isn’t afraid to copy Samsung’s smaller Z Flip phones, which flip down horizontally, and larger devices like Lenovo’s X1 Fold and ASUS’s ZenBook Fold 17. An “iPhone Flip” could potentially satisfy people demanding a new iPhone Mini, and a future foldable iPad Pro could truly differentiate that tablet from the increasingly capable iPad Air.

This all seems obvious, of course. Now that Apple has shown it can develop ultra-thin device connected by a sturdy hinge, it should be easier for the company to scale its designs smaller and larger. I’m just hoping it doesn’t take forever. After all, it took Apple nearly seven years to finally respond to the Galaxy Z Fold (which, admittedly, was practically a beta product for its first few generations). What better way for Apple’s new CEO, John Ternus, to follow up the iPhone Duo’s debut than with a smaller iPhone Flip next year? And given how both ASUS and Lenovo have been slow to improve their large foldable tablets, there’s room for Apple to truly innovate in that space.

Sam Rutherford for Engadget

What would an “iPhone Flip” look like? Apple wouldn’t have to stray far from the template laid out by Samsung and others. Take an iPhone 17, then imagine casually folding the top down, just like how we used to close older cellphones. Also like Samsung’s devices, an iPhone Flip would need a small external display to glance at notifications and handle basic tasks, like taking a selfie with its rear cameras.

Such a design wouldn’t be an exotic innovation for Apple. Small foldables have been around for six years now; they’ve been practically normalized. They typically cost half as much as large foldable smartphones (and in many cases even less). Anecdotally, I’ve also noticed many “normie” smartphone users toting the Z Flip or Moto Razr, whereas I rarely see the Z Fold being used by non-techies.

While Apple is rumored to be working on a major iPhone redesign for its 20th anniversary next year, I’m not sure a bezel-less screen with curved edges will really excite anyone. But an iPhone Flip that can easily slip into a pocket or small purse? I’ve been begging for that for years.

Sam Rutherford for Engadget

When it comes to a foldable iPad, Apple also has room to do far more than ASUS or Lenovo, both of which are stuck building for Windows. Based on what we’ve seen of the iPhone Duo, Apple has thought deeply about how to juggle apps and functionality across folding screens. As usual, it has the software edge. An iPad Fold could replicate the look of the iPhone Duo by having a smaller iPad Mini-sized screen up front, and a larger screen internally. But I could also see Apple forgoing the external display entirely for cost reasons. Personally, I’d want a device like this for its fullscreen tablet and folded, pseudo-laptop modes.

Sam Rutherford for Engadget

When fully opened, it could work just like any iPad today. But imagine folding it in half and using the top half as a laptop screen, while the bottom is reserved for a touchscreen keyboard with haptic feedback. Or you could use that folded formation to multitask: Play a YouTube video up top, while juggling between Slack and hand-written notes on the bottom half. I could also imagine having it stand up vertically while still folded in half, giving you taller windows for multi-tasking. Slap on a wireless keyboard, and it’s practically like having a super-light mini-desktop

Both ASUS and Lenovo relied on large 16.3-inch OLED screens for their foldable tablets, but I don’t think Apple needs to go that large. A 13-inch foldable iPad Pro would still be pretty useful when cut in half. A smaller 11-inch iPad Fold could be a great portable journal, similar to the unreleased Surface Neo. (And yes, it’s hilarious that Apple now has products using the Duo and Neo names, as if it was rubbing salt into Microsoft’s wounds.)

Figuring out the intricacies of a foldable iPad would likely take Apple years, and truthfully, there’s not a huge demand for something like that just yet. But Samsung has done much of the groundwork for delivering a potential iPhone Flip, Apple just has to follow that template.

As AI platforms become part of daily workflows, attackers have found a new way in: the platforms themselves. The Huntress Security Operations Center (SOC) says the bigger day-to-day risk comes from threat actors abusing the AI features people already trust and rely on, rather than attacks on the AI companies or models themselves.

Over the past nine months, Huntress has tracked incidents in which attackers weaponized shareable AI content, public mini-apps, and sponsored search placement to target AI users and deliver malware.

Legitimate features, hijacked

Huntress has observed threat actors abuse a handful of real AI platform features, including:

• Claude Artifacts: content Claude generates and displays in a chat preview pane, which users can publish and share via a public link.

Advertisement

• claude.ai/share links: shareable URLs created when someone publishes a Claude conversation; these can surface in search engines when posted to crawlable spots like forums or social media.

• ChatGPT and Grok conversations: shared, indexable conversations hosted on chatgpt.com and grok.com that can rank for troubleshooting searches.

Each of these sits inside a trust boundary. Users recognize the platform, the branding, and the surrounding content, so malicious instructions or downloads look legitimate. These campaigns often only run for hours or days before a provider pulls the content down, but that’s enough time to trick victims before getting caught.

FakeAgent: malvertising through a Claude Artifact

In July, Huntress saw a campaign called FakeAgent hit more than 29 organizations. It started with a malicious Claude Artifact hosted on the real claude.ai domain.

Since public Artifacts are meant for lightweight demos and get minimal vetting from Anthropic beyond a generic disclaimer, attackers built a convincing fake Claude Desktop download page.

Victims searching Bing for the Claude desktop app landed on the fake page and clicked what looked like a legitimate download link. Instead, they were redirected to an external domain that delivered the SectopRAT malware.

Huntress reported the Artifact and Anthropic removed it by July 22, but incidents tied to the same redirect domain continued into August.

A fake install guide hiding in claude.ai/share

In a separate incident, a victim searching Google for “Claude on Mac” clicked a sponsored result that led to a claude.ai/share link posing as an Apple Support install guide. Because the page lived on Anthropic’s own domain, it carried none of the usual red flags: no lookalike URL, no certificate warning.

The fake guide instructed the victim to paste a curl command into Terminal, kicking off a six-stage chain that deployed the MacSync stealer. It harvested cookies, credentials, keychain secrets, Telegram sessions, and SSH and cloud keys.

AI poisoning via ChatGPT and Grok

A third pattern targets AI-generated troubleshooting advice itself. In December, a routine search for “clear disk space on macOS” surfaced high-ranking ChatGPT and Grok conversations that gave ClickFix-style instructions instead of real fixes.

Attackers had crafted the conversations, hit “share” to generate a public URL on the platform’s trusted domain, and used SEO poisoning to push the link to the top of Google’s results.

Because the links lived on real chatgpt.com and grok.com domains, victims trusted the advice and ran the suggested Terminal commands, which delivered the AMOS stealer. 

What defenders should do

None of these attacks broke through the AI platform security. They exploited the trust users place in familiar brands and real domains. 

Defenders should treat clipboard-driven execution and AI-assisted troubleshooting as security risks. Restrict script execution from the clipboard and enforce application allow-listing. Watch for new scheduled tasks and antivirus exclusion changes, and train users to spot ClickFix-style lures. Report suspicious AI-hosted content to the platform vendor quickly.

These campaigns tend to be short-lived, but fast reporting and layered controls can shrink the window attackers get to exploit them.

If you’re interested in this kind of tradecraft and exploring how attackers evolve their tactics, join our experts at Tradecraft Tuesday, where we break it all down every month. 

Sponsored and written by Huntress Labs.

For the people behind the attacks, ClickFix now makes their job much easier. Prior to ClickFix, they would have needed to install the malware (tracked as Lorem Ipsum, security firm BlueVoyant said recently) using resource-intensive infrastructure, including SEO-manipulated and malvertised download portals, Microsoft-trusted signing certificates, and continuously rotated domains for delivering Microsoft Installer packages.

“The pivot to ClickFix in late May 2026 eliminates the code-signing requirement entirely, substituting the legitimacy of a validly signed installer with a different form of legitimacy: a user voluntarily executing the malicious command in their own terminal,” BlueVoyant said. “[T]he ClickFix model broadens the victim pool from users searching specifically for Microsoft Teams to anyone browsing a compromised website.”

The situation for macOS users isn’t any better. Both Mac security firm Jamf and a researcher have ​​documented macOS variations of ClickFix that can bypass Gatekeeper protections.

ClickFix attackers keep finding new ways to use public services—including publicly published Google Sheets documents, according to Cisco Talos. Other attackers, including Russia’s state-sponsored Sandworm, are hosting their control infrastructure in blockchain-based smart contracts. Security firm Netskope recently found another campaign that used the same approach. The security company counted 5,400 sites beaconing to it, an indication of the reach and scope of that campaign. And as OS makers and defenders build new defenses, attackers keep finding documented ways to work around them.

The upshot of all this is that ClickFix is a highly effective and efficient means of spreading all sorts of malware. It’s not going away, and victim-blaming or shaming only makes the problem worse.

There are a fair number of plugins, standalone products, and built-in defenses that are designed to blunt the success of ClickFix attacks. For instance, BlockBlock, the software that monitors Macs for processes that seek to permanently install themselves, can block ClickFix attacks as soon as a user presses the ⌘+V keys. Ublock has been updated to do something similar.

Beyond those fixes, those of us with more security training should build awareness with our less experienced neighbors, family members, and friends. The mass adoption of ClickFix demonstrates its success, and it’s not going away any time soon.