In brief: It’s hard to miss a drone. With their spinning propellers and often bulky bodies, they’re usually very easy to spot. But Northwestern University engineers have created one that’s much more difficult to see coming: Phantom Twist, a prototype that rotates so quickly it becomes little more than a blurry haze.

Rather than using camouflage, transparent panels, or Predator-style light-bending materials, Phantom Twist exploits motion blur and the limits of human vision.

Its body spins between 15 and 25 times per second, causing the eye to average the moving parts with whatever lies behind them. At maximum speed, the aircraft is nearly invisible, appearing as a faint smudge.

Unlike a conventional quadcopter, which has four rotors surrounding a mostly stationary body, Phantom Twist uses a single motor and propeller.

The propeller rotates in one direction while the rest of the aircraft spins the other way, leaving almost no stationary components for the eye to lock onto.

Creating something that could remain stable while also being difficult to see required more than attaching components to a rotating frame. The researchers began with a computational model that generated around 20,000 viable configurations. AI and optimization algorithms then rearranged the motor, batteries, circuit board, counterweight, and other parts to find designs that balanced flight performance with low visibility.

The most promising candidates were simulated while spinning and overlaid across 100 real-world backgrounds. A perception model designed to approximate human vision scored how noticeable each one appeared. The team retained the 500 least visible designs and continued adjusting their layouts until it arrived at the final version.

Its widely separated components sit at different heights and angles, with plenty of empty space between them. This prevents the solid pieces from overlapping during rotation and forming a dark, recognizable shape. According to the researchers’ visibility metric, Phantom Twist is around 10 times less perceptible than a similarly sized quadcopter.

There are a few caveats. The current prototype relies on an external optical tracking system, limiting it to controlled environments, while its exposed wires and carbon-fiber support rods remain partially visible. More importantly, it still sounds like a drone, so anyone who fails to spot the hovering blur will probably hear it. Future versions could use quieter propulsion and more transparent materials.

The drone’s potential applications include observing wildlife without altering animal behavior, surveying wetlands, and inspecting infrastructure with less visual disruption. A camera mounted on the spinning body might also capture a 360-degree view for navigation.

Similar aircraft designed to be difficult to detect include China’s mosquito-sized surveillance drone and Harvard’s solar-powered RoboBee. Phantom Twist takes a different approach to the same goal: rather than shrinking out of sight, it moves too fast to see clearly.

Last year’s trip to CanJam London was something of a nightmare, with lengthy train delays and torrential rain pouring over the London Eye and Westminster Bridge.

This time, the journey into the heart of the city could not have been easier. There were no travel hiccups, and the sun managed to find its way through every available gap between the buildings, buses and cars.

Inside the Park Plaza Hotel, the air conditioning was working overtime, which was just as well. Attendance was clearly up compared with 2025, and there were noticeably more brands and exhibitors. So many, in fact, that I simply could not see everything on offer over the two days.

Following our recent coverage of CanJam NYC 2026, and with the eCoustics team heading to CanJam SoCal at the end of the summer, London offered another useful snapshot of where personal audio is heading. With my trusty Sendy Audio Egret in tow, I focused on the products that stood out most. We begin with my personal highlights among the DACs, headphone amplifiers and DAPs at CanJam London 2026.

Related Reading:

Burson Audio Conductor Stellar

Australian audio brand Burson has seemingly come out with a smaller, more compact Conductor Voyager that costs less and offers almost the same level of performance.

You get two watts less of Class A amplification, at eight watts rather than ten, but you won’t notice that unless you’re using a particularly difficult-to-drive pair of cans. What’s more, the Stellar has a dedicated IEM amplifier stage, so you can drive your sensitive in-ears with ease.

Despite being Class A, I found the Stellar to be wholly neutral, albeit with excellent dynamics and an impactful bass line. As a pure headphone listener, I’d go for this instead of the Voyager unless you need the extra I/O that the latter offers at the back.

Fosi Audio ZH1

I’ve always been a fan of Fosi Audio’s budget-conscious approach to personal audio, and its upcoming ZH1 certainly sticks to that philosophy.

Expected to arrive around September for approximately £130, the ZH1 is a compact desktop DAC and headphone amplifier built around a pair of Cirrus Logic CS43131 DAC chips and Fosi Audio’s own headphone amplifier circuit. Its unusual hexagonal chassis also helps it stand out from the endless procession of small black boxes competing for space on your desk.

That oversized volume knob definitely takes centre stage. It makes quick adjustments easy, while the lighting underneath can be changed to suit your setup. Some audiophiles will undoubtedly switch it off immediately, but I imagine the colourful presentation will make the ZH1 particularly popular with gamers and younger listeners.

This is not merely a piece of affordable desktop decoration, either. With more than one watt of output power available, the ZH1 should have enough muscle for a broad range of headphones rather than being limited to sensitive IEMs and easy-to-drive models. It certainly had no difficulty with my Sendy Audio Egret at the show.

Aune A17

Next up is this absolute behemoth of an amplifier. I’ve reviewed a few of Aune’s products in the past, but nothing has come close to the sheer heft of this beast, which weighs an astonishing 18 kg.

It has the power to back up its size, too, delivering almost 12 watts from its balanced output. That translates rather directly into its sonic performance, as my main takeaway was the exceptionally high level of control it maintained across the entire audible spectrum.

It also costs upwards of £2,000 ($2,399 at Amazon), and while you certainly get a lot of material and componentry for the money, I found the design rather uninspiring and, dare I say it, slightly cheap-looking.

FiiO R7 R2R

The FiiO R7 R2R is expected to cost the same as its original sibling at around $700, but replaces the standard R7’s ESS ES9068AS delta-sigma DAC architecture with an R2R resistor-ladder conversion stage.

The rest of the formula remains familiar. It is still a compact, all-in-one desktop streamer, DAC, preamplifier and headphone amplifier, with a vertical touchscreen, Android-based access to streaming services and support for local playback. The standard R7 also offers Wi-Fi, Bluetooth, AirPlay and Roon Ready connectivity, alongside balanced 4.4 mm and four-pin XLR headphone outputs and a 6.35 mm single-ended connection.

To me, the move to R2R represents a clear upgrade. I noticed the difference immediately, with greater bass impact, more warmth and a fuller overall presentation than I heard from the standard R7.

I love that FiiO continues to refine and expand its range of all-in-one streamers, DACs and headphone amplifiers rather than abandoning successful products after a single generation. The R7 R2R does not reinvent the original, but it sounds like another worthwhile incremental improvement to one of FiiO’s most versatile desktop components.

FiiO Class A

Shuffling further down the FiiO booth, I spotted another as-yet-unreleased product: a dedicated headphone amplifier called the CLASS A. The name rather gives the game away.

This is a fully discrete, pure Class A design that delivers one watt per channel. That might not sound particularly dramatic on paper, but it sounded considerably more powerful than the figures suggested, driving my headphones with plenty of authority and control.

Inside is a custom 60-watt toroidal transformer and a discrete regulated power supply, while five selectable gain levels should make it suitable for a wide range of headphones. Outputs include 3.5 mm, balanced 4.4 mm and four-pin XLR connections, with both RCA and balanced XLR line-level connectivity also provided.

More importantly, it delivered the warmth and sweetness I tend to associate with Class A amplification without sounding overly soft or sluggish. I was also a fan of the physical bass and treble controls, which make it easy to fine-tune the sound without opening an app or disappearing into a menu. You do not see enough of those nowadays.

FiiO has yet to confirm the final international release date or pricing. I was quoted a provisional Chinese price of approximately 3,000 RMB, which converts to around $450, although taxes, distribution costs and regional pricing will inevitably have their say.

Even allowing for that, the CLASS A looks like excellent value for money.

Sendy Audio Kylin

I listened to the Sendy Audio Kylin at last year’s show and was so impressed that I named it Best in Show in the midrange price category.

One year later, my impressions have not changed one jot. Some of its specifications might not look especially impressive on paper, particularly beside amplifiers boasting increasingly ridiculous power figures, but the Kylin sounds absolutely gorgeous.

Vocals and guitars were rich, full-bodied and wonderfully expressive, while the presentation remained spacious, punchy and highly dynamic. There was enough warmth to make everything sound inviting, but not so much that the music became soft or overly relaxed.

Under the bonnet, the Kylin combines an XMOS USB stage with an ESS ES9038Q2M DAC, native DSD512 decoding and a pure Class A headphone amplifier using matched Hitachi transistors. It also offers balanced 4.4 mm and four-pin XLR headphone outputs, alongside a 6.35 mm single-ended connection.

None of that fully explains why I enjoyed listening to it so much, which rather proves the point. Numbers are useful, but they are not everything in the world of audio.

Astell&Kern A&Ultima SP4000T

The Astell&Kern SP4000T is, very loosely speaking, an SP4000 with four Raytheon JAN6418 vacuum tubes bolted on. There is rather more going on beneath the surface, but the description gets the general point across.

Those four military-specification tubes are arranged in a differential configuration, with two dedicated to each channel. The SP4000T also lets you choose between Triode, Ultra Linear and Pentode tube modes, along with OP AMP, Tube AMP and adjustable Hybrid amplification. Factor in three selectable tube-current settings, and there are plenty of ways to tailor its presentation.

It costs an eye-watering £3,200 and weighs an equally shocking 565 grams, making the word “portable” feel rather generous. You are not slipping this into the pocket of your jeans unless you would also like to explain the resulting wardrobe malfunction.

You do, however, get your money’s worth sonically. The SP4000T delivered a bombastic, commanding presentation that, to my ears, holds its own against many desktop DAC and headphone amplifier stacks costing similar money. There was plenty of scale and low-end authority, but it never sounded uncontrolled or clumsy.

The rest of the package feels suitably premium, with a smooth interface, responsive touchscreen and full Android support for direct access to streaming apps. Anyone seeking the best of the best in portable DAPs will undoubtedly consider it, although at 565 grams, they may also want to consider a stronger bag.

Fosi Audio DAP Neo

Back at the Fosi Audio booth, I was introduced to another prototype on display: the DAP Neo.

This is a deliberately simple, offline digital audio player. It does not connect to the internet, which means no Spotify, Apple Music, TIDAL or other streaming services. You load your music onto the player and listen to it. That may sound rather old-fashioned in 2026, but sometimes removing notifications, apps and endless choices is exactly the point. It allows you to focus on the music and nothing else.

The compact design looks more than a little like one of Apple’s classic iPods, which is hardly the worst place to draw inspiration from. It felt straightforward to operate and refreshingly uncomplicated beside the increasingly large and expensive Android players found elsewhere at the show.

Fosi expects the DAP Neo to cost less than $100, although the final price and specifications have yet to be confirmed. It sounded solid during my brief demonstration and powered my Sendy Audio Egret comfortably, with enough weight and clarity to make listening enjoyable.

At this price, I am not asking for a desktop-class powerhouse or a miniature smartphone. Give me reliable local playback, a usable interface and enough power for a sensible pair of headphones, and there is not much more I could ask for.

Best Portable DAC/Amp: Audma Brioso PHPA1

Admittedly, the Audma Brioso PHPA1 was the only portable DAC and headphone amplifier I listened to at CanJam London this year, but that does not make it any less deserving of our Best in Show award.

Audma is based in the same region of Italy that my father comes from, and the company is responsible for developing its patented ELISA circuitry. The technology is designed to recreate some of the spatial characteristics of listening to a pair of loudspeakers, including the small timing differences between the sound reaching each ear.

The Brioso lets you adjust the perceived distance and width of the presentation using its Stage and Angle controls. It is also a serious portable DAC and amplifier in its own right, using AKM’s flagship AK4499EXEQ DAC alongside the AK4191EQ digital filter and modulator. It supports USB-C digital playback, balanced and unbalanced analog inputs and up to five hours of battery-powered listening.

I tried the Brioso with ELISA switched both on and off, and the difference was immediately noticeable. With it engaged, the soundstage appeared to move in front of me rather than extending mainly to either side of my head. The effect was not subtle, and I still find it rather extraordinary that carefully controlled differences in timing and spatial information can change the presentation so convincingly.

The fact that Audma has managed to achieve this inside a genuinely portable device makes the Brioso even more impressive.

Mind you, it costs almost £4,400, so it probably should be.

Best DAP: FiiO M25 R2R

As we recently reported, FiiO brought two forthcoming DAPs to CanJam London: the M25 and M25 R2R. Positioned below the flagship M27 and above the more affordable M33 R2R and M21, the new models should help bridge what has become a rather substantial gap in FiiO’s portable player lineup.

The standard M25 uses AKM’s flagship separated-DAC architecture, combining an AK4191 digital filter with two AK4499EX DACs. The M25 R2R takes a very different approach, replacing those chips with the second generation of FiiO’s proprietary R2R PRO resistor-ladder technology.

Both versions use Qualcomm’s Dragonwing QCS6490 processor, supported by 8GB of RAM and 128GB of internal storage. They also include a 31-band parametric equalizer, Auto EQ and FiiO’s Desktop Mode, which allows the player to draw power from an external supply rather than continually cycling the internal battery.

I listened to the resistor-ladder version, and it struck a fine balance between naturalness and sweetness without sacrificing clarity. Vocals and instruments had a fuller, more organic quality, but the presentation never became overly soft or syrupy.

My only minor gripe was the absence of a physical volume wheel. FiiO has opted for a conventional rocker instead, which works perfectly well, but perhaps I am simply one of those people who still prefers turning a proper dial.

Both M25 models remain unreleased, with final international pricing and availability yet to be confirmed.

Best Budget DAC: Topping E50 II

At the Electromod booth was the Topping E50 II, the Chinese manufacturer’s revision of its well-regarded E50 DAC. Priced at $199 at Apos Audio and Amazon ( £199 or €199), depending on the market, I am astounded by how much functionality Topping has squeezed into such a small and affordable unit.

The E50 II uses AKM’s AK4497S DAC chip alongside an XMOS XU316 USB interface, with support for PCM playback up to 32-bit/768kHz and native DSD512. Inputs include USB-C, coaxial, optical and Bluetooth, with LDAC, aptX HD and aptX Adaptive support. It also offers RCA and balanced 6.35 mm TRS outputs, which can operate simultaneously, and can be used as either a fixed-output DAC or a digital preamplifier.

The star of the show is its parametric EQ, which can be configured through the Topping Home app and stored on the unit for offline use. That is a considerable amount of flexibility for a $199 DAC, particularly for listeners who want to correct a headphone’s tonal balance or make smaller adjustments to suit their system and personal preferences.

In terms of listening, the E50 II has a no-nonsense, neutral sound signature, but with just enough musicality to prevent it from sounding flat, sterile or boring. It does not attempt to impose much of its own personality, which is exactly what many listeners will want from an affordable desktop DAC.

For $199, what’s not to like?

Best Midrange DAC/Amp: Luxsin B9

This was one product I was not expecting to see in London. For all intents and purposes, the Luxsin B9 is an R2R interpretation of the X9, although this was my first experience with any of the company’s products.

The engineering sample combines Luxsin’s proprietary discrete R2R DAC architecture with a headphone amplifier and preamplifier in a single desktop unit. Luxsin has not published the complete production specifications yet, but the B9 is expected to deliver up to six watts of headphone output power and cost around $1,500 when it is released.

The prototype also included some genuinely clever features. Luxsin’s HP-EQ system lets you select a correction profile for your headphones, adjust it to suit your own preferences and save the results. There is also a hearing compensation function that tests the sensitivity of each ear across different frequencies before creating separate correction curves for the left and right channels.

I was not able to test either feature properly at the show, but the concept is far more useful than adding another digital filter that most owners will try once and never touch again.

The B9 is also beautifully designed, with sharply angled casework and a bright, vibrant touchscreen that makes most competing desktop DACs look rather ordinary. More importantly, its R2R presentation was exciting and expressive without becoming aggressive or fatiguing. It made a considerable impression on me during a relatively brief listen.

For a product I was not expecting to see, the Luxsin B9 became one of the more memorable electronics demonstrations of the weekend.

Best High-end Amp: MASS-Kobo Model 491

And finally, we come to MASS-Kobo, a Japanese audio brand whose amplifiers are hand-built and assembled to order in Shizuoka under the direction of Masuda-san. I was glad to spend a few minutes brushing up on my Japanese with him and the rest of the team at their booth.

The Model 491 ($20,000 USD) is the successor to the flagship Model 465 ($19,000 USD) and looks very similar at first glance, with the most obvious external change being the addition of a 4.4 mm balanced headphone output. It sits alongside four-pin XLR, separate three-pin XLR outputs for the left and right channels, and a 6.35 mm single-ended connection.

Inside, the 491 uses a twin-balanced monoblock design with four amplifier stages, two for each channel, and fully independent left and right power supplies. It delivers at least 3.5 watts per channel into 32 ohms through its balanced outputs, which should be enough for even the most stubborn planar magnetic headphones.

Being the flagship amplifier from a boutique high-end brand, it costs a barely believable £16,599 through Elise Audio. Most of that money is going into the circuit design, component selection and the highly trained ears of Mr. Masuda, rather than a touchscreen, app support or an unnecessarily complicated menu system. The amplifier itself is sturdily built, but it does not offer much in the way of bells and whistles.

And my goodness, does the 491 sound stunning.

I tend to prefer more neutral-sounding equipment, and the latest MASS-Kobo certainly falls into that category. However, it walks the line between analytical precision and musicality more successfully than almost any headphone amplifier I have heard. Sounds appeared to drift in and out of thin air, as though there were nothing standing between the music and my ears.

It came remarkably close to creating the impression of listening to a live performance rather than a pair of headphones. I arrived with very high expectations, yet the Model 491 managed to exceed them anyway.

The Bottom Line

CanJam London 2026 made one thing very clear: personal audio has never offered more choice. From sub-$100 offline DAPs and $200 DACs to £16,599 hand-built amplifiers, the show covered almost every conceivable use case and price point.

I could not hear everything, but the products above were the ones that made me stop, listen and return for another track. Roll on CanJam London 2027.

Related Reviews:

Apple is expanding its presence in Sunnyvale once again by leasing a 125,800-square-foot building. If history repeats itself, Apple will buy it within a year.

Companies with a massive workforce are always on the lookout for more office space. In the case of Apple, it will be adding one more location to its roster in Sunnyvale.

According to sources of the Mercury News, Apple has agreed to lease a building located at 580 North Mary Avenue in Sunnyvale. The property, which weighs in at 125,800 square foot of office space, is being leased from the Bay Area real estate firm Peery Arrillaga.

The terms of the lease have not been leaked yet, but it will certainly cost Apple millions per year to occupy.

A building buying spree

While the property is being leased for the moment, Apple has the potential to buy it at a later time. This has become a habit of the company, choosing to lease before acquiring it fully.

Apple’s June purchase of office space in June demonstrates this perfectly.

After leasing the office building at 684 W. Maude Avenue in Peery Park, Sunnyvale in 2025, Apple waited until later in the year to close the deal to purchase it. That 194,624 square foot property cost Apple $162.2 million, which was a considerable discount from its $222 million sale price in 2022.

Its Silicon Valley real estate moves in the area has also included a 2025 acquisition of a two-building campus in North Mathilda for $350 million. That was closely followed by another four-building Mathilda campus just next door for $365 million.

The same year, Apple paid $160 million for a 220,700-square-foot building in the Tantau office complex, close to Apple Park.

Kodak just handed film beginners a camera they can actually stick in a jeans pocket without second thoughts. This little plastic box called the EC35 costs thirty-five dollars and works with any standard roll of 35mm film. Reto Project builds it under the Kodak name, and the whole point is simple: give people who have never touched film something that feels as easy as the disposable cameras they grew up seeing at drugstores, only this one lasts longer than a single roll.

A sliding cover lies on top of the lens and viewfinder, ready to be pushed open to prepare the camera for action. Sliding it shut closes the shutter, allowing you to put the camera in a bag or pocket without fear of scratching the glass or accidently taking a frame. This haptic little detail is evocative of the good old Olympus Stylus days, and it’s ideal for flinging this little number around without fear of harm. The body is a compact device that measures about four and a half inches across and weighs only 3.5 ounces. It comes in seven various colors: midnight black, vanilla white, butter yellow, lavender, powder blue, blush pink, and avocado green.

The camera itself is about as simple as it gets; there’s no fumbling around with settings for new users, as the lens is fixed at 25mm and f/10 with no ability to modify it, and the shutter fires at a constant 1/100 of second. There is an automated flash tucked down near to the lens, powered by a single AA battery, that turns on when the light goes out. A simple manual film advance and rewind, similar to the old reliable point and shoots, allows you to insert a 24 or 36 frame roll of film (which must be manually wound by the way) and shoot your way through it with a little of twiddling on a convenient small window above the lens and pushing a button.

If you want to get started with the camera, a starter kit that includes a neck strap and a roll of Kodak Ultramax 400 film costs $45 dollars. Alternatively, you can buy the camera body separately for $35. In any case, the pricing is a deal, significantly lower than any of the other film cameras on the market today with all the bells and whistles. There is no complex zone focusing, double exposure lever, or aperture dial, since the camera simply compels you to work with what the fixed lens and shutter provide, much like in the disposable days.

Because the body is reusable, the only ongoing cost is having extra film developed, which is much less expensive than purchasing a new plastic camera every time you wish to take a few dozen photos. You can acquire one of these small devices right now from the Kodak store, although early shipments of the more vibrant hues have already sold out. For $35 and a single AA battery, it’s a wholly unique alternative to a cameraphone.

from the lying-people-into-life-sentences dept

Record law enforcement officers whenever you can. Sure, they’ll hate it, but it helps all the people they seek to do harm to.

Last summer, recordings and other evidence undercut a lot of the administration’s attempts to convert First Amendment activity into criminal charges. Prosecutors who dropped cases after watching recordings or talking directly to the government’s witnesses noted a plethora of “untruthful statements” or “material misrepresentations.”

Six months later, the losing streak created by unforced (but deliberate) “errors” continued. Federal immigration officers just couldn’t stop lying, forcing DOJ prosecutors to voluntarily dismiss a bunch of felony assault cases that were completely reliant on statements made by officers who were later shown to be lying.

Here’s another case that’s hit a dead end because an ICE officer lied and an ICE supervisor decided to pass on these lies to a federal prosecutor. (h/t Aaron Reichlin-Melnick)

A Venezuelan man arrested by ICE was hit with a slew of escalating charges, most of which have vanished now that the truth is involved.

Gabriel Hurtado-Cariaco, 31, was referred to by DHS as a “criminal illegal alien” and a “known Tren de Aragua terrorist.” The agency said he “violently attacked an ICE agent” by slamming her head into the ground and attempting to choke her to death. At a preliminary hearing, a Homeland Security agent and a federal prosecutor said Hurtado-Cariaco received “special forces-type training” in the Venezuelan military and may have been involved in “killing people, hurting people, [and] committing human rights violations.”

The only thing true here is that Hurtado was once a member of the Venezuela military. But he deserted, fled to the US, and filed an asylum claim. He was paroled into the US in 2024 and, since then, had been steadily working as a delivery driver.

The arrest of Hurtado didn’t look great. Two ICE officers tried (and failed) to subdue him. He managed to escape custody, but was arrested successfully shortly after he fled the scene of the first attempted detention.

Perhaps the (attempted) arresting officer was too embarrassed to tell the truth. But the lies he told had serious consequences.

The initial narrative of what happened during Hurtado-Cariaco’s arrest was provided by Immigration and Customs Enforcement agent Craig Allrich. Allrich, who was not at the scene of the arrest, penned the affidavit that would be used to secure a criminal complaint against Hurtado-Cariaco.

The first complaint actually told the truth. It simply said that Hurtado had resisted arrest. But that apparently wasn’t good enough for Allrich and/or the officer who failed to subdue Hurtado during his first encounter with him. Things got worse for Hurtado in a hurry.

It was in the amended complaint that Allrich laid out how Hurtado-Cariaco allegedly attempted to murder the female ICE officer, referred to as “Victim 1” throughout the document. 

[…]

[A]s the two agents struggled to detain Hurtado-Cariaco, he was “able to get control of Victim 1 and place [her] in a chokehold.” Allrich said the male agent gave repeated commands to Hurtado-Cariaco to release the chokehold, but he “chose to continue choking Victim 1 rather than simply fleeing the scene.”

Advertisement

This assertion was a lie. Two recordings made by bystanders who witnessed the attempted arrest made it clear Hurtado had definitely resisted arrest, but at no point did he pin down the female ICE officer, nor did he ever place her in a chokehold.

At no point in either video is Hurtado-Cariaco seen choking the female agent. Instead, the videos show agents placing Hurtado-Cariaco into a chokehold on two separate occasions as he attempts to break free.

Hurtado’s federal public defender, Richard McWilliams, says federal prosecutors sat on these recordings and refused to remove the attempted murder charge despite having evidence proving otherwise. US Attorney Lesley Woods says this isn’t true, claiming the public defender refused to turn over the recordings he had obtained.

But Woods’ claims are difficult to believe, especially when prosecutors waited months to rewrite the indictment, excising the bogus attempted murder charge just so that they could replace it something nearly as damaging.

By November, prosecutors filed a superseding indictment, which abandoned the attempted murder charge and replaced it with another serious felony – providing material support to a terrorist organization.

According to prosecutors, the mere existence of a tattoo — one that doesn’t actually demonstrate a connection to Venezuelan gang Tren de Aragua — was enough to hit Hurtado with terrorism charges.

(And that charge is only possible because the Trump administration unilaterally declared this gang to be foreign terrorist organization — something it only did in an attempt to legalize its unconstitutional detain-and-deport program.)

None of this bullshit impressed the judge handling Hurtado’s criminal case.

U.S. District Court Chief Judge Robert Rossiter said at the sentencing that the allegations made in the initial complaint were “at worst a misrepresentation and at best complete negligence.” Though Rossiter said there is “no doubt” that Hurtado-Cariaco committed the offense that he pleaded guilty to by fleeing the officers and struggling with them, he found the law enforcement reports “embellished” and “troubling.”

That’s a polite way of saying either the ICE officer at the scene lied or that the ICE supervisor who prepared the affidavit didn’t care whether or not he was being lied to.

And the judge isn’t happy with the government’s last ditch attempt to stick Hurtado with a lengthy sentence by belatedly adding some terrorism charges to the mix. As it stands now, Hurtado’s guilty plea to resisting arrest nets him a 14-month sentence, most of which he has already served because he’s been in jail since his arrest last June.

“To refer to him as a terrorist, attempted murderer, that he had put a chokehold on these victims… It’s just not borne out by the evidence,” Rossiter said. “And it’s troubling.”

If you can’t do your job without lying, you can’t be trusted to hold that position. If the administration can’t find enough people to arrest and deport without having to make up stories about murder or terrorism, then it needs to restrain itself to the promise that has always been empty: go after the worst of the worst. These lies were caught. Dozens or hundreds of others will go undetected.

Given what we know about the people “leading” this country and these agencies, I firmly believe they’d lie even if they didn’t have to. That’s just the kind of people they are. And the foot soldiers willingly serving this rolling atrocity of an administration need to be filmed whenever possible, because it’s the only way to keep them honest.

Filed Under: craig allrich, ice, mass deportation, rights violations, robert rossiter, trump administration, venezuela

A single AI agent conversation can look flawless scored on its own and still point to a broken product. That gap is driving a shift in how enterprises evaluate agents, away from scoring individual traces and toward comparing cohorts of users against a baseline.

At VB Transform 2026, Harrison Chase, CEO of LangChain; Hui Zhang, CTO and co-founder of Conviva; and Emmanuel Turlay, director of engineering at CoreWeave, described that shift, along with a parallel move toward cheaper, narrower judge models.

Agent-as-judge — judging one AI agent’s output with another — hasn’t replaced LLM-as-judge, which Chase said remains the default. The larger tension, Zhang said, is between automated judging, whether by LLM or agent, and human review.

“You have scalable but ungrounded, whether it’s agents as judge or LLMs as judge, you grade the outcome, you grade the work. It still is very difficult to ground it and then you use humans and that’s just not scalable,” Zhang said. “The whole industry is facing this, which poison you want to pick.”

Evaluation criteria now function as the product spec

That gap — a conversation that scores well but still signals a broken product — is what teams try to close by building an exhaustive evaluation suite before they ship anything. Chase said that doesn’t work.

“We sometimes see teams that have almost eval paralysis,” Chase said. “They’re like, this is an eval set, I can’t launch it. The best teams launch and then iterate.”

Chase framed evaluation criteria as a living specification, not a one-time test suite: a product requirements document — the standard software-development spec for what an application should do. “Evals are like the new PRD,” he said. “They define what your agent should and shouldn’t do.”

Turlay described hitting the same failure from a different angle. “I was trying to reach 100% coverage for my tests, and I still had bugs in production,” he said — a test suite that looked complete but still missed what mattered, the same gap Chase was describing with evals.

Broad, always-on monitoring, he said, catches more real failures than an exhaustive pre-launch test suite. Teams should set up wide online checks first, use those to identify failure classes as they occur, then build a targeted offline evaluation set around the problems that surface.

Why scoring traces one at a time is a mistake

Even a well-built evaluation process can still score the wrong thing. Zhang’s objection is to how most teams run evaluation: sampling traces, whether 50 of them or a full population, scoring each in isolation. That approach misses a signal that only shows up when comparing cohorts of users against a baseline, a method Zhang calls contrastive analysis.

Zhang illustrated it with a retail example: a shopper asks an agent for a running shoe ahead of a half marathon, the agent asks qualifying questions, and the shopper buys a shoe. Scored individually, that interaction looks fine. But the clarification ratio, how many follow-up questions an agent asks before completing a task, came in three times higher than baseline for that shoe category across the full user population. A second metric, how often shoppers finished their purchase outside the conversation, was five times higher than baseline for the same category.

Neither number is visible from a single trace. Both point to a debuggable, category-specific problem. Zhang said the industry also lacks a second data source: what happens before, between and after the conversation, not just the trace itself.

Sizing the judge to the job

Once contrastive analysis flags which category is actually broken, the next problem is what watches for it going forward — and at what cost. Turlay’s rule was to start with the most capable model available to prove a task is solvable, then work down. If it can’t be done with a top-tier model, he said, it won’t work with a smaller one. Once a pattern proves viable, teams can sample a fraction of traffic instead of judging every interaction, and move simpler tasks like binary classification to smaller open source models.

LangChain took that further, fine-tuning its own model to detect when a user believes the agent made a mistake, a signal Chase calls perceived error. “The model we fine-tuned was a Qwen model,” he said, referring to Alibaba’s open source family. Combining hand labeling with distillation, the result performed well. “Same as [Claude]Sonnet, for, depending on how we served it, either 10 to 100x cost reduction,” Chase said.

Not every guardrail needs a model. Chase pointed to Claude Code’s own guardrails as proof: regexes, the common programming technique for finding and validating patterns in code. “A lot of the guardrails they had were just regexes,” he said. “They weren’t small LLMs, they were just regexes.”

LLM-as-judge doesn’t mean human-in-the-loop disappears

The bigger question is whether using LLM as a judge removes the need for a human in the loop.

Turlay pointed to accountability, drawing on his prior work at a self-driving car company. His team compressed data intake and retraining into a two-week cycle for shipping a new model to the car. Even then, someone still had to sign off.

“I felt confident on behalf of the company to say this model should go into the car,” he said. The same logic extends to legal, finance and healthcare. “Before we can remove a human to say, I endorse this and I take responsibility legally for it, it’s going to be a while before agents can do that on their own.”

Zhang agreed a human has to remain the guardian on corner cases, even as automation eventually runs at a scale that beats individual human accuracy — machines can see more at the pattern level.

Chase went further: that human check isn’t just a safety net. “Human in the loop is really important for building trust in how these agentic systems work, and also really important for memory and learning from systems,” he said. “There has to be interactions in order for the system to learn.”

Robert Downey Jr. returns to the Marvel Cinematic Universe this morning in a role no one saw coming a few years ago. He wears the metal mask of Doctor Doom in the first full trailer for Avengers Doomsday, and the footage makes clear this is no simple villain introduction. Directed by Anthony and Joe Russo, the film lands in theaters on December 18 and pulls together heroes from three separate universes who suddenly share the same battlefield.

Patrick Stewart’s Professor X appears in the teaser with a hushed warning, a sense that something big is about to happen that even a seasoned veteran like him may be unable to stop. By the end of the day, he warns everyone that they’ll have to make a decision they can’t even think about, and that sentence hits home since X has seen enough multiverse mayhem to know when the normal techniques won’t cut it.

Sale

LEGO Marvel Avengers: Endgame Final Battle – Super Hero Building Set for Kids, Ages 10+ – 10 Figures…

• BRING MARVEL STUDIOS’ AVENGERS: ENDGAME TO LIFE – This LEGO Marvel Avengers kids’ building toy is an authentically detailed set for kids to…
• LEGO AVENGERS MINIFIGURES – Captain America, Black Panther, Iron Man, The Scarlet Witch, Ant-Man, Iron Spider, Doctor Strange, Falcon and Chitauri…
• MARVEL ACTION FIGURES – The heroes burst through magic portals into an epic battle with Thanos as kids recreate an iconic movie scene and role-play…

Then Chris Hemsworth’s Thor appears, delivering the trailer’s longest and most urgent speech, attempting to convey his message. He recalls men far stronger than everyone currently standing who died because the threats they faced were more terrifying to him than the one they confront today. He goes on to warn that every single sacrifice those guys made will be in vain until the current group can put aside their old feuds and come together as brothers and sisters. Then he musters the courage to state unequivocally that they will require nothing less than a miracle.

That miracle appears in the form of Chris Evans as Steve Rogers. After years on the sidelines, Captain America returns to the frame, reaches out, and lifts Thor’s hammer as if it were nothing. Thor’s astonished reaction demonstrates how much time has gone since Rogers’ departure and how little some aspects of him have truly changed.

We also get our first good look at Downey’s Dr. Doom in full armor. Then Thor smacks him with a burst of energy, leaving him immobilized. He’s already three steps ahead of any plan you can think of. That sequence underscores Doom’s dominance over all other adversaries in the MCU.

As the trailer progresses, more and more characters from different universes cross paths, with Pedro Pascal, Vanessa Kirby, Joseph Quinn, and Ebon Moss-Bachrach appearing as the Fantastic Four and encountering people that do not belong in their reality. Then there are the original X-Men – Patrick Stewart, Ian McKellen, James Marsden, Rebecca Romijn, Alan Cumming, Kelsey Grammer, and Channing Tatum – who appear alongside Anthony Mackie’s Captain America, Sebastian Stan’s Bucky Barnes, Florence Pugh’s Yelena Belova, Paul Rudd’s Ant-Man, and so on.

Tickets for the exclusive Infinity Vision screenings went on sale the same morning the trailer debuted. This is the final chapter of the Multiverse Saga before Secret Wars, and it appears to be a doozy. The trailer doesn’t explain how these three universes merged or what Doom wants to get out of it; all it does is show the scope of the problem and the unusual group of people who are now standing in his way.

Microsoft is working to fix a known issue affecting Windows Server Update Services (WSUS) servers, which has caused synchronization problems for more than a week.

WSUS was introduced almost twenty years ago to help IT administrators schedule updates for Microsoft products on enterprise networks from a single local update server, rather than updating each endpoint directly from Redmond’s servers.

By default, WSUS syncs with Microsoft Update servers once a day to download the latest metadata for available Windows updates. However, on servers affected by this issue, synchronization is broken, and admins will not be able to deploy the latest Windows updates via WSUS or Configuration Manager.

According to Microsoft, the WSUS sync issues affect both client (Windows 10, version 1607 and later) and server (Windows Server 2012 and later) platforms.

“Organizations might experience increased synchronization times or sync operation timeouts on WSUS servers. This issue began in recent days, with heightened impact observed starting July 13, 2026,” Microsoft said in a Windows health dashboard update.

Microsoft has rolled out mitigation measures on Saturday to prevent WSUS sync times and operation issues from affecting newly installed or rebuilt WSUS servers.

However, it is still working on additional mitigations for previously affected WSUS servers that continue to experience WSUS synchronization issues.

“Synchronization times and sync operations on WSUS servers have been restored and are operating normally for new WSUS installations and rebuilds,” it added. “For WSUS servers that were previously affected, Microsoft is working on mitigation steps to help customers safely remove the affected metadata from their environments.”

One year ago, in May, Microsoft fixed another similar issue after enterprise customers reported being plagued by WSUS errors when trying to update Windows 11 22H2/23H2 systems.

Microsoft addressed another WSUS issue in July 2025 that prevented organizations from syncing with Microsoft Update to deploy the latest Windows updates.

One month later, Microsoft resolved one more issue that blocked the August 2025 security update from being delivered via Windows Server Update Services (WSUS).

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

A malicious component dubbed HollowGraph uses the calendar feature in compromised Microsoft 365 mailboxes as a command-and-control channel to receive attacker commands and exfiltrate stolen data.

Researchers analysing the module believe it is part of the Cavern command-and-control framework that has been previously linked to an Iranian threat actor targeting entities in Israel.

At least 12 systems have been infected with HollowGraph, three of them actively communicating with the threat actor between June 3 and July 9.

The collected indicators suggest that the threat actor is focused on organizations in Israel, pointing to a targeted attack for espionage purposes.

Microsoft 365 mailbox abuse

In a report from cybersecurity company Group-IB, researchers say that HollowGraph uses hardcoded details to authenticate to the Microsoft Graph API via a compromised Microsoft 365 account.

The configuration file is stored as logAzure.txt to appear as a regular log file, and “includes the Microsoft Entra ID tenant ID, application (client) ID, client secret, target mailbox address, command-and-control (C2) domain, and two RSA keys.”

The two cryptographic keys are used to encrypt files before delivery to the attacker and to decrypt incoming tasks.

To remain under the radar, the threat actor creates calendar events dated May 13, 2050, with the title in specific formats. Commands and exfiltrated data are concealed within files attached to these calendar entries.

According to Group-IB’s analysis, HollowGraph supports only two commands that let it create calendar entries with stolen files in encrypted form and search for new ones with instructions from the threat actor:

• GET, to search for entries in the format “Event ID: <7-char-taskID>,” download and decrypt received instructions
• SEND, to create a calendar entry in the format “Boss{..}ID{..}” and attach stolen data encrypted with the public RSA key

The researchers describe the mailbox calendar as a “covert dead-drop,” with HollowGraph retrieving commands from events scheduled within a fixed one-hour window between 22:00 and 23:00 UTC on May 13, 2050.

Group-IB explains that the threat actor uses a hybrid encryption scheme that mixes RSA and AES-256-GCM algorithms to secure the communication over Microsoft Graph, keeping inbound and outbound channels cryptographically separated.

HollowGraph has a second, unencrypted communication channel through DNS tunnelling, which is used to receive new Microsoft Entra ID details (Entra ID: tenantId, clientId, clientSecret, and mailbox) to authenticate to Microsoft Graph.

It retrieves the values through IPv6 AAAA record queries to the attacker-controlled domain cloudlanecdn[.]com, and updates the configuration files stored as logAzure.txt.

“Each returned IPv6 address (16 bytes) yields 14 usable payload bytes,” Group-IB explains. The malware assembles the payload from these 14-byte chunks, decodes it as UTF-8 text, and stores the result according to the corresponding configuration field.

While the researchers cannot attribute HollowGraph to a known threat actor, their “analysis identified several technical similarities with the Iranian-nexus threat actor Lyceum.”

However, the available evidence is insufficient to attribute the activity to the threat actor with high confidence. By contrast, the researchers assess with high confidence that HollowGraph is linked to the Cavern framework.

Group-IB suggests that organizations monitor Microsoft Graph and Microsoft 365 audit logs for suspicious application-driven calendar activity, particularly events in the far future, and unusual subjects and attachments.

It is also recommended to look for indicators such as the ‘cloudlanecdn[.]com’ domain and the ‘logAzure.txt’ file, enforce Conditional Access, restrict and audit OAuth client-credential applications, and monitor outbound DNS for tunneling patterns.

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