Job recruiters are meeting an uncanny new kind of candidate: the AI avatar. These digital humans mirror an applicant’s voice and likeness while being something else entirely, and they are turning up in first-round interviews.

Lana Kersanava, a Sydney-based recruiter, told Semafor she has encountered them over the past six months. “At the beginning, it’s like you feel like something is off,” she said, adding: “It looks very realistic.”

How she spotted it

The tell was not the rendering, which held up. It was the answers, delivered with a polish that never faltered across a run of questions.

After roughly five minutes, Kersanava ended the interview and disqualified the candidate. She has now done this three times.

The detail worth pausing on

All three avatar candidates had applied for English-speaking roles as non-native speakers, leaning on the technology to cover a gap in language proficiency. That is a different proposition from someone faking a work history.

It also lands in a hiring market with a documented bias problem. Research has found recruiters up to 19% less likely to follow up with job seekers from immigrant and ethnic minority backgrounds than with equally qualified candidates.

None of which makes an undisclosed avatar acceptable. It does complicate the framing of these candidates as straightforward fraudsters.

The system they are responding to

Applicant tracking systems already use AI to filter CVs before a human reads them, and recruiters increasingly deploy AI avatars to conduct first-round interviews themselves. Candidates are now answering machines with machines.

The resulting AI-on-AI war is rewriting the hiring playbook, pushing employers towards live technical walk-throughs, scenario challenges, and roleplay simulations that are harder to automate through.

The volumes explain the arms race. Nearly two-thirds of candidates now use AI in applications and 87% of US hiring managers use it in hiring, alongside a 239% surge in AI-generated applications.

How common this is

A 2026 Greenhouse report found 91% of US hiring managers have encountered or suspected AI-generated answers during online interviews. Across 19,368 live interviews analysed between July 2025 and January 2026, 38.5% of candidates were flagged for AI-assisted behaviour.

The trajectory points one way. Gartner projects that by 2028 one in four job candidate profiles globally will be fake, spanning AI-generated profiles, deepfake video interviews, and fabricated work histories.

Barriers to entry are minimal. Palo Alto Networks found it takes as little as 70 minutes for someone with no image manipulation experience to build a fake candidate capable of passing a video interview.

Employers are retreating to the room

Seventy-two per cent of recruiting leaders now run at least one in-person stage specifically to counter AI-assisted fraud, with Google, McKinsey, and Cisco reintroducing mandatory face-to-face rounds.

That solves detection at the cost of everything remote hiring was meant to deliver. Candidates who cannot travel, including many of the international applicants this technology most tempts, lose access first.

The screening was never that good

There is an awkward premise underneath the panic, which is that the automated filtering being defended works. The case against AI CV screening has been made repeatedly and forcefully, largely on the grounds that it optimises for keyword matching rather than capability.

Candidates optimising against that filter with better tools is a predictable outcome rather than a moral collapse. Kersanava’s five-minute detection rate also suggests the current generation of avatars is not fooling attentive humans.

What it is doing is making the first round worthless for everyone. If the machine screens the machine, the interview stops measuring anything, which is an argument for changing the process rather than policing the candidates.

If you purchased an iPhone 16 or iPhone 15 when they were released in order to jump into promised Apple Intelligence features, you were probably disappointed: The signature AI abilities were nowhere to be found, and only a few related features appeared later.

The good news is that you may be entitled to receive a portion of a quarter-billion-dollar lawsuit. The bad news is that you’ll have to wait more than a year to see any of it.

Apple settled a false advertising lawsuit in May, agreeing to pay $250 million to customers who bought the iPhone 16 and some iPhone 15 models during a specified period. The lawsuit alleged that Apple misled customers by promising AI features that didn’t ship when the new devices did. Payouts between $25 and $95 per eligible device are expected.

In an earlier statement to CNET Managing Editor David Lumb, an Apple spokesperson said, “Apple has reached a settlement to resolve claims related to the availability of two additional features. We resolved this matter to stay focused on doing what we do best, delivering the most innovative products and services to our users.”

Now the case has moved forward as of July 17. According to an email from Clarkson Law Firm, “The Court has granted preliminary approval of the settlement … This means the Court has reviewed the proposed terms and found them sufficient to move forward to the next phase; it is not a final ruling on the settlement’s fairness.”

Why is there a lawsuit over Apple Intelligence?

When Apple advertised its new iPhone 16 lineup in 2024, it emphasized how the models were optimized for AI features such as an enhanced Siri that could act as an intelligent agent. When the phones did arrive, Apple Intelligence wasn’t yet ready; its first features didn’t arrive until iOS 18.1, five weeks later.

According to the proposed settlement, “Apple allegedly saturated the market with deceptive ads, inducing consumers to purchase iPhones based on the promise of certain enhanced Siri features.”

Some features of Apple Intelligence did ship soon after the introduction of the iPhone 16 and iOS 18, including Visual Intelligence, Live Translation, Writing Tools, Genmoji and Clean Up. But those weren’t the advanced features Apple highlighted. In fact, the type of intelligent interactions Apple advertised are only starting to show up in Siri AI, which is part of iOS 27 that’s expected to be released later this year. The public beta of iOS 27 is available now.

Who is eligible for the $250 million settlement?

Customers who purchased one of the following devices between June 10, 2024, and March 29, 2025, are eligible to receive a settlement payment:

• iPhone 16
• iPhone 16E
• iPhone 16 Plus
• iPhone 16 Pro
• iPhone 16 Pro Max
• iPhone 15 Pro
• iPhone 15 Pro Max

The iPhone 15 Pro and iPhone 15 Pro Max are included because they had the processor and memory to run Apple Intelligence features.

It’s estimated that approximately 36 million customers are eligible for this settlement.

How to claim your portion of the settlement

For now, you need to wait.

As set forth in the settlement, Apple will provide a list of eligible customers and their contact information to a settlement administrator. 

After the data has been verified, the company Verita will send email and postal notices to those customers directing them to a settlement website. That site has not yet been created. The deadline for filing your claim will be 90 days after your notice arrives.

When can you expect to receive a settlement payment?

According to the settlement, Apple needed to provide the information about affected customers within five days of the settlement approval. As of late July, preliminary approval has been granted by the court.

After the data is provided and verified, a 45-day notice period begins to inform potential consumers that they’re eligible for a payment.

According to the July 17 motion, the final approval hearing is scheduled for the morning of Sept. 29, 2027.

Provided approval is granted, that puts the first checks or deposits arriving sometime within a 60-calendar-day window after September 2027, depending on court dates and possible extensions.

Jeff Carlson

Senior Writer

Jeff Carlson writes about mobile technology at CNET, from cellular phone plans to devices and emerging technology.

See full bio

Good morning! Let’s play Connections, the NYT’s clever word game that challenges you to group answers in various categories. It can be tough, so read on if you need Connections hints.

What should you do once you’ve finished? Why, play some more word games of course. I’ve also got daily Strands hints and answers and Quordle hints and answers articles if you need help for those too, while Marc’s Wordle today page covers the original viral word game.

Programming with Claude Code will soon require even less human oversight, as Anthropic says it’s making auto mode the default for Pro, Max, and Team accounts, starting on August 14.

The company first unveiled a test version of auto mode in March, pitching it as a way to balance speed and control. As Anthropic explained in its announcement on Friday, when Claude Code is in auto mode, instead of presenting prompts asking for human approval at each step, it will proceed unless an action is determined to be “irreversible, destructive, or aimed outside your environment.”

Anthropic also said that in testing, auto mode proved safer than manual review — in a study with 1,053 paid testers, auto mode caught 89% of harmful actions, while human review only caught 13.6%. (Perhaps that’s because “manual review can become habitual: users approve 97% of permission prompts in Claude Code.”)

In a post on X, Claude Code Head Boris Cherny said, “The team and I use Auto mode exclusively, and have been for many months. I couldn’t imagine going back to permission prompts!”

The company also said it’s been adding new safety features like prompt injection screening and customizable hard deny rules to prevent things like data exfiltration.

And why you might need to.

What’s the best storage format for Windows (and other devices you own)?

Max Miller for Engadget

How to properly format your storage drive for Windows

Max Miller for Engadget

High-end audio does not have a sound-quality problem. It has a discovery problem.

Too much of the industry’s best equipment is demonstrated at audio shows, dealer events and private listening sessions to people who already understand why someone might spend five figures on a turntable, tube amplifier or pair of horn loudspeakers. Those events matter, but you can only preach to the converted for so long before somebody notices that most of the congregation already owns three turntables and has very strong opinions about 12AX7 tubes.

Which is why what Devon Turnbull and OJAS are doing at the Cooper Hewitt, Smithsonian Design Museum in New York deserves considerably more attention from the high-end audio industry.

Turnbull’s HiFi Pursuit Listening Room Dream No. 3 places a large-scale, handmade OJAS audio system inside Cooper Hewitt’s historic Carnegie Library, where visitors can actually sit down and listen rather than stare at an amplifier sitting silently behind a velvet rope. The installation originally opened in December 2025 as part of the museum’s Art of Noise programming and has now been extended through August 16, 2026.

This Is About More Than OJAS

The significance is not simply that Turnbull has managed to get a high-end audio system inside a Smithsonian museum. It is who gets exposed to it once it is there.

Art of Noise explores more than a century of music and design through over 300 objects and installations, including record players, radios, album artwork, the Sony Walkman, Apple iPod and experimental contemporary audio designs. Those are cultural objects that already mean something to people who may never have attended AXPONA, HIGH END Vienna or CanJam and who probably could not tell you what a moving-coil cartridge is if their life depended on it.

But they understand music. They understand design. Many of them buy vinyl, go to concerts, care about architecture and furniture, collect art, own good headphones or obsess over the way their homes look.

That audience should be fertile ground for high-end audio, yet the industry has historically done an impressive job of making itself difficult to discover.

OJAS is doing the opposite.

A Listening Room Built Around the Space

Dream No. 3 is the third version of Turnbull’s listening-room concept, following earlier installations at Lisson Gallery in New York and London and SFMOMA. For Cooper Hewitt, however, Turnbull designed the system specifically around the museum’s historic Carnegie Library, including custom speaker sculptures influenced by the room’s existing millwork. Cooper Hewitt describes it as the largest and most architecturally and acoustically integrated version of the installation so far.

That matters because the system is not being presented like another pile of expensive components.

It becomes part of the room.

Visitors are encouraged to sit, stop talking, put away the phone and actually listen. The room is programmed every day with selections ranging from classical and ambient music to jazz, while Turnbull, members of the OJAS team, musicians, engineers, collectors and other guest operators have conducted live listening sessions throughout the exhibition.

Turnbull returns for live operator sessions on August 13 and August 15 from 1:00 p.m. to 4:00 p.m., just before the installation closes on August 16.

High End Audio Needs More of This

We have been following Turnbull and OJAS for years, from his collaboration with Denon on the DL-103o moving-coil cartridge to the Klipsch kO-R1 and newer kO-R2 loudspeakers. The Klipsch partnership has been particularly interesting because the products have deliberately appeared outside the normal hi-fi circuit, including Milan Design Week, where the kO-R2 was presented to an audience that crosses architecture, furniture, fashion and industrial design.

We praised that decision at the time, and Cooper Hewitt reinforces why.

High-end audio companies keep asking how they attract younger customers and people outside the existing audiophile community. Perhaps the answer is not another seminar about Ethernet switches in Room 614 of an airport hotel.

Put great systems in museums. Put them in architecture and design exhibitions. Build listening rooms inside record stores, galleries, restaurants and cultural spaces. Partner with musicians, furniture designers and people who already have credibility outside the audiophile bubble. Most importantly, let people hear what a genuinely good system can do without first requiring them to identify as an “audiophile.”

The industry sometimes forgets that almost nobody wakes up one morning and decides they need a $20,000 pair of loudspeakers. They hear something that changes their perception of reproduced music, become curious and start moving down the rabbit hole.

Some eventually spend $500. Some spend $5,000. A few lose all fiscal judgment and start pricing Swiss monoblock amplifiers.

But somebody has to open the door.

The Bottom Line

HiFi Pursuit Listening Room Dream No. 3 is worth celebrating not simply because Devon Turnbull built another visually distinctive OJAS system, but because Cooper Hewitt has placed serious high-fidelity reproduction inside a broader conversation about music, culture and design.

That exposes high-end audio to people who might otherwise never experience it, and the industry needs considerably more of that.

Audio shows and specialist dealers remain essential, but they predominantly reach consumers who have already walked through the door. If high-end audio really wants to expand its audience, it needs to start appearing in places where curious music lovers are already standing.

OJAS figured that out.

Everyone else might want to take notes.

For more information: cooperhewitt.org

Related Reading:

The first images of a SpaceX Falcon 9 rocket stage impact on the moon’s surface have emerged.

The Korea Aerospace Research Institute shared the first photographs captured by its lunar orbiter. South Korea is one of the few nations that currently has satellites traveling around the moon at an altitude of 62 miles (100 kilometers) above the surface. The Danuri orbiter flew over the region shortly after the impact.

KARI explained on its X account that it photographed the collision site both before and after the event. The images show signs of a recent impact, which astronomers projected would likely create a crater dozens of feet wide nearly 12 feet (3.6 meters) deep.

This evidence of the collision adds to observations from the Very Large Telescope, where astronomers detected sodium and lithium rising from the lunar surface, confirming that the rocket fragment did indeed strike the moon.

The crater is larger than several cars combined.

Photograph: Korea Aerospace Research Institute

Astronomers predicted when and where the Falcon 9 impact would happen well in advance, and there were signs it could be visible with sufficiently powerful equipment. Yet days before the collision, space agencies such as NASA warned that it would be difficult to observe the moment of impact due to weather conditions and timing.

Like KARI, NASA has lunar orbiters. It’s possible that in the coming days, following a series of data checks, higher-resolution images of the impact area will become available.

But even at higher resolution, getting a granular view of the crater is unlikely. In fact, the new photos give a sense of just how small that is compared to the vastness of the lunar surface. To put this into perspective, the moon measures about 2,160 miles in diameter.

Another image taken by the Danuri lunar satellite during one of its orbits over the impact area.

Photograph: Korea Aerospace Research Institute

Why Did the Rocket Crash?

The Falcon 9 fragment made a crash landing after being in an unstable orbit for more than a year. The rocket launched in January 2025, completing its mission to deliver lunar landers. For more than a year, the gravity of Earth, the moon, and the sun slowly altered its orbital trajectory until it eventually impacted the lunar surface.

“For most of our missions, we plan for a controlled deorbiting of the Falcon second stage so that it re-enters safely over the ocean,” SpaceX said of the rocket. “In higher-energy missions, such as those heading to a lunar transfer orbit, nearly all of the vehicle’s performance is dedicated to successfully placing the payload into the intended orbit, and a controlled deorbiting maneuver is not always possible.

“We are actively working to be as responsible as possible with the hardware left in space and to ensure space safety, even on more complex missions. In this case, over time, solar activity and gravity deflected the second stage toward the Moon. Impacts like this are rare, but they can occur with objects in these types of orbits, and we are collaborating with NASA to find the optimal disposal solution,” it added.

This story originally appeared on WIRED en Español and has been translated from Spanish.

David Silver is a billionaire, sort of. His stake in Ineffable Intelligence—the artificial intelligence lab he founded—is worth an ungodly fortune. But he stands to collect none of that money himself. He set things up that way.

Between 2013 and 2025, Silver worked as a researcher at Google DeepMind. He specialized in reinforcement learning, a method of teaching an AI model through positive and negative feedback. In 2016, he led the development of AlphaGo, which became the first AI to defeat a human champion in the game of Go and was heralded as a narrow form of superintelligence. Silver left to start Ineffable Intelligence in January, betting that reinforcement learning breakthroughs will close the gap between human and machine cognition across the board.

When he started Ineffable Intelligence, Silver raised $1.1 billion from investors at a $5.1 billion valuation, reportedly the largest seed round in Europe’s history. He also vowed to donate all the money he makes from any future sale of his company to charity.

“I was thinking very deeply about what I could do to have the maximum impact on the world,” Silver tells WIRED in an interview at his office in Kings Cross, London, around the corner from DeepMind’s headquarters. “Not just through the work I do, but through the finance I arrive at in the process.”

He’ll choose how to deploy the funds after he sells his company, he says, but his ambition is to save as many lives as possible.

Silver is the latest in a growing line of AI founders to pledge equity proceeds to charity. Others include Mustafa Suleyman, a DeepMind cofounder and the CEO of AI at Microsoft, and Anton Osika, the founder of automated coding platform Lovable. If OpenAI and Anthropic go public as planned, minting thousands of new millionaires in the process, the AI industry could soon come to represent an increasingly outsize proportion of tech philanthropy.

Along with a new generation of techno-philanthropists, though, comes a tangle of unresolved questions: Is philanthropy the most appropriate way to redistribute wealth? To what degree are megadonors responsible for some of the problems they aim to fix? Could a pledge that scales with a company’s success come to justify unsavory business practices or reckless accelerationism? And is this go-around any different from previous iterations of Silicon Valley giving?

“You can’t necessarily appeal to capitalism to solve some of the problems created by capitalist practices themselves … the exacerbation of extreme wealth concentration and massive inequalities,” says Linsey McGoey, professor of sociology at the University of Essex and author of No Such Thing as a Free Gift. “It’s like calling upon the arsonist to hose down the house he’s just set on fire.”

Silver formalized his charitable commitment through Founders Pledge, a nonprofit that helps entrepreneurs to give away their wealth. Unlike the Giving Pledge, a public but nonbinding commitment, Founders Pledge requires participants to enter into a contract. “The choice of making a binding pledge was very important,” says Silver.

Founders Pledge has been around since 2015. But the value of pledges recorded by the nonprofit has soared recently, from $400 million in 2023 to $4 billion so far this year. The AI industry now accounts for a third of the lifetime pledge total.

The organization’s goal, according to cofounder David Goldberg, is to identify “the very best ways to deploy capital at scale to make the world better.” It has frequently been lumped together with effective altruism, which encourages people to amass wealth in order to give it away, to rely on statistics rather than emotion, and to prioritize causes likely to have the biggest impact in the distant future. But despite the commonalities, Goldberg isn’t a fan of the comparison; Founders Pledge is concerned with human suffering in the present. “We’re far more humanist,” he says.

Something to look forward to: Engineers at the University of Texas at Austin have developed a soft patch that helps people enter REM sleep faster and stay in that stage longer, without drugs or surgery. In a study involving 28 people, participants who used the patch, called NEUSLeeP, reached REM sleep 43 minutes earlier than usual and remained in that stage for about 16 minutes longer.

The device works by combining two things that don’t usually go together in a wearable: gentle ultrasound stimulation and electrodes that read brain activity. This allows it to reach deep brain regions associated with REM sleep while also tracking the brain’s response in real time from outside the skull.

Kai Wing “Kevin” Tang, the recent UT biomedical engineering Ph.D. graduate who led the work, said the achievement, which was published in Nature Communications, marks new territory for the field.

Researchers have never before been able to noninvasively reach deep brain regions involved in REM sleep while simultaneously monitoring brain activity as it occurs. According to Huiliang “Evan” Wang, an assistant professor in UT’s Department of Biomedical Engineering who supervised the research, the combination of stimulation and real-time monitoring gives the device its practical edge. He noted that a skin-attached device such as NEUSLeeP could give researchers and doctors new ways to study sleep and treat sleep disorders in patients’ homes.

The benefits weren’t limited to sleep timing. In healthy participants, the stimulation also increased heart rate variability, a measure used to assess how well the body responds to stress. Brain scans showed changes in circuits associated with emotion, suggesting that the patch could do more than help people sleep – it might also help them cope with stress.

Gregory Fonzo, an assistant professor in the Dell Medical School’s Department of Psychiatry and Behavioral Sciences and a co-principal investigator, tied these findings to the role REM sleep plays in the brain. He explained that REM isn’t just the stage when dreaming happens – it is also when the brain processes emotions and adapts to stress. As a result, boosting REM sleep could potentially give people a new tool for coping with stress and improving overall well-being.

That’s significant because disruptions in REM sleep are repeatedly associated with depression, anxiety, and PTSD. Current treatments, such as medication and therapy, don’t always address underlying sleep problems and can come with side effects.

Both healthy sleepers and people with existing sleep problems responded well to the patch in the study. Participants described it as comfortable, and researchers reported few side effects.

The researchers plan to conduct larger trials to confirm the findings and determine whether the patch can help people with PTSD, depression, and chronic insomnia specifically. They also believe it could be useful for at-home sleep monitoring and neuroscience research, since noninvasive tools that can simultaneously stimulate and record brain activity are relatively uncommon.

The researchers are working with Discovery to Impact, UT’s commercialization office, to bring the patch closer to market. A patent application has already been filed.