ai and ml

San Francisco plays host to hosting company’s Localhost conference

On Thursday during a developer event held at San Francisco’s St. Regis Hotel, the marketing stack failed. Literally.

During an AI agent workflow demo on the fourth floor balcony, the wind toppled a tower of three cardboard promotional display cubes. One of these – maybe three or four feet per side – fell over the edge of the balcony and plummeted onto Minna Street below.

Concerned staff rushed to the edge of the parapet and peered over. Evidently satisfied that no one had been injured, they proceeded to dismantle the other stack of boxes, just to be safe – an uncommon level of caution in the context of AI-assisted software development. 

Consider the incident a metaphor for the chaos created by the tech industry’s frantic rush to automate whatever can be automated with machine learning models and associated tools.

The event, Localhost, was put on by hosting biz Render as a confab for devs building software for the “AI-native web.” But many attendees expressed uncertainty about the rapid pace of change in the industry.

As one CTO attendee remarked, not realizing he was in earshot of a reporter, “How’re we going to do this? I don’t fucking know. That’s what I have to figure out.”

The CTO said he was looking for engineers to hire, the very thing major tech companies have been dumping to balance capex costs.

Rohan Chavan, who recently earned his master’s degree in computer engineering from Virginia Tech, shared a similar sentiment.

“Every other day,” he told The Register in an interview, “there’s some new term. A week or two ago, it was harness engineering. Now it’s loop engineering.”

Chavan said he found the current state of play both exciting and frustrating. He’s looking for a job in AI security and estimated that about 30 percent of his class of around 120 has been hired so far.

It’s very difficult, he said, to build deep knowledge when things are changing so fast and there’s so much to learn.

Localhost, the company’s first user conference in its eight years of existence, aims to help with that by evangelizing the company’s technical stack. 

Some promotion might just be useful. As the aforementioned CTO confided to one of Render’s staff, enterprise customers expect to hear that services run on AWS, Azure, or maybe Google Cloud. “They don’t expect to hear Render,” he lamented.

Nonetheless, Render is doing rather well, according to founder and CEO Anurag Goel, who opened the afternoon’s presentations with the requisite recitation of metrics – 400,000 developers joining every month, 10 million live services, 200 billion monthly requests.

“All of these numbers are growing very rapidly,” he said. “But behind these numbers, we’re also seeing a big change in how applications are evolving to use infrastructure.”

Until a few years ago, he said, infrastructure was relatively static. You deployed an application and the various elements – databases, APIs, caches, and so on – stayed pretty much the same.

AI apps, he contends, are fundamentally different.

“They are dynamic applications that go beyond existing infrastructure patterns,” Goel explained. “Now, in addition to you defining your application statically, the applications themselves are provisioning their own infrastructure resources.”

As an example, he described how the resources required by a research agent can vary from question to question. One user query might require lightweight web scripting. Another might require a headless browser and 128 GB of RAM to process a dataset. And the duration of such tasks can also vary significantly.

“What we’re learning is that any request can trigger hundreds or thousands of different tasks in ways that your code can never really know in advance,” he said.

This doesn’t work on serverless platforms, Goel contends, because of platform limitations that cap execution time, memory, storage, and application size.

Render’s answer to this technical challenge is what Goel calls application-defined compute. It allows applications to run workloads without pre-provisioning. 

“Instead of pre-defining the infrastructure for your application, you allow the application to define what it needs at runtime with the right guardrails,” he explained.

With any luck, Render’s guardrails will prevent a pummeling by promotional props better than those at the St. Regis. ®

A New York man faces cyberstalking charges after allegedly sharing AI-generated nude images and fabricated racist messages using fake social media profiles to harass a Georgia college student.

21-year-old Anthony Belford was arraigned June 10 after a federal grand jury returned an indictment charging him with one count of cyberstalking.

Belford and the victim had attended the same college during the 2023-2024 academic year. After the victim transferred to a Georgia college in August 2024, Belford allegedly knew of the move and began targeting the victim there.

According to court documents, between January and March 2025, the defendant created fake Instagram, LinkedIn, Reddit, X, Strava, and Yahoo accounts to impersonate the victim and distribute AI-generated nude images and spread false claims that the victim had made racist remarks about black students and anti-Muslim statements.

Belford allegedly created a fake LinkedIn profile using an AI-generated nude image of the victim as its profile picture, and also used a spoofed Yahoo email account to send an AI-generated nude image of the victim to the victim’s mother.

The defendant allegedly targeted the victim while attending the same college in the 2023-2024 academic year, but continued doing it even after the victim transferred to a Georgia college in August 2024.

“Belford allegedly waged a lengthy online campaign, hiding behind spoofed social media and email accounts to harass, intimidate, and cause substantial distress to his victim with racist messages and AI-generated nude images,” said U.S. Attorney Theodore S. Hertzberg.

“Cyberstalking and other forms of online abuse, just like physical violence, can ruin lives and disrupt communities. Victims of such crimes should not suffer in silence, and we will continue to work with our law enforcement partners to hold the perpetrators of these crimes accountable using all available tools.”

The Justice Department added that federal law prohibits sharing or threatening to share intimate images (including AI-generated ones) without consent and urged victims to report violations to the FBI and to alert the Federal Trade Commission if online platforms fail to remove such content within 48 hours of a removal request.

More information on how to protect yourself from cyberstalking attempts and stop the spread of images and videos shared online without consent is available on the FTC’s Take It Down platform.

In March, 22-year-old Jamarcus Mosley from Alabama also pleaded guilty to cyberstalking, extortion, and computer fraud charges after hacking into the social media accounts of hundreds of young women.

The same month, 26-year-old Kyle Svara from Illinois also pleaded guilty to hacking nearly 600 women’s Snapchat accounts to steal private nude photos that were later traded or sold online.

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.

Advertisement

Get the whitepaper

The bottleneck in the AI buildout is turning out to be a metal most people have never heard of. China has tightened its scrutiny of exports of indium phosphide, a compound essential to the high-speed optical chips that move data inside AI data centres, in a move that threatens to slow the very infrastructure the technology depends on.

Indium phosphide, or InP, is not a household material, but it is becoming a strategic one. As data-centre operators shift from pushing electrical signals through copper to sending light through optical fibres, a technique known as photonics, InP has become the core material with no ready substitute.

The faster the AI industry wants to move data between chips, the more it needs the compound, and China sits at the chokepoint.

That position is a matter of geology and processing. China produces around 70% of the world’s indium, and since export controls on InP took effect in early 2025, Beijing has been slow to approve the licences that let the material leave the country.

The delays, rather than an outright ban, are the lever: a permit that does not arrive is as effective as a prohibition, and harder to challenge.

The market has felt it. The price of a six-inch InP wafer has climbed from roughly $1,400 to about $5,000 since the controls began, an increase of around 250%, as buyers compete for constrained supply.

Nvidia-backed chipmaker Coherent warned of a shortage earlier this year, and AXT, the world’s second-largest InP substrate producer, has described the export permits as the most significant challenge it currently faces.

The episode fits a now-familiar pattern in the US-China technology contest. Where Washington has restricted China’s access to advanced chips and chipmaking tools, Beijing has answered by leveraging its dominance over critical materials, having already deployed controls on gallium, germanium, and rare earths.

InP is the same weapon pointed at a different part of the supply chain, the optical layer rather than the logic layer.

What makes InP potent is precisely that it targets infrastructure rather than end products. The compound goes into the transceivers and optical components that knit together the thousands of accelerators in a modern AI cluster, so a squeeze on it does not stop any single chip from working; it slows the rate at which whole data centres can be built and wired. The constraint shows up as delayed construction, not failed silicon.

It also lands as the AI industry’s appetite for compute is at its most acute, with operators racing to build capacity faster than the supply chain can support.

The same pressure visible in the scramble for chips and components now extends to a niche material that few outside the industry tracked a year ago. China’s leverage over it has turned a specialist input into a geopolitical instrument.

The deeper worry for the AI industry is precedent. If a delay in InP permits can slow data-centre construction, the same lever can be applied to any of the specialised inputs where China holds a commanding share, turning a diversified supply chain into a series of single points of failure.

That fragility is now a strategic planning problem for Western governments and operators alike, part of the wider contest over technology supremacy in which materials have become as decisive as the chips they enable.

Substitution offers little near-term relief. Building InP production capacity outside China is possible but slow, requiring years of investment in refining and wafer fabrication that the current shortage does nothing to accelerate.

In the meantime, buyers are left managing allocation, paying the higher prices, and lobbying through diplomatic channels for the permits to move, a position of dependence that the controls were designed to exploit.

The InP controls were also raised directly with Beijing; Coherent’s chief executive brought the licensing delays up during a US business delegation’s visit to China, a sign of how seriously the buyers take the threat.

Whether the permits start flowing again, and on what terms, is now part of the broader negotiation between the two governments over technology and trade. For the AI buildout, the answer determines how fast the lights can go on.

OFFBEAT

Tribunal rejects bid to strike blacklisting claims, with proceedings due to conclude shortly before GTA VI launches

Rockstar Games has suffered a legal setback in a dispute over alleged union busting, clearing the way for a final employment tribunal hearing shortly before Grand Theft Auto VI is due to launch.

The developer had sought to have “blacklisting” allegations struck from the case. The employment tribunal rejected the request, and the final hearing is scheduled to run from September 10 to October 15.

According to the Independent Workers’ Union of Great Britain (IWGB), which brought the case, blacklisting is a practice in which information about workers engaged in union activity is compiled to facilitate discrimination.

The Register asked the IWGB for more information, but the union did not respond. Rockstar declined to comment on ongoing legal matters.

The legal dispute stems from the sudden dismissal of 31 IWGB members in October 2025. According to the IWGB, the dismissed workers were part of a private trade union Discord channel where they discussed ways to improve the workplace.

An anonymous source told The Register that when management became aware of the channel, the staff were summarily fired.

At the time, a spokesperson for Rockstar’s holding company, Take-Two Interactive, said: “We strive to make the world’s best entertainment properties by giving our best-in-class creative teams positive work environments and ongoing career opportunities. Our culture is focused on teamwork, excellence, and kindness.

“Rockstar Games terminated a small number of individuals for gross misconduct, and for no other reason. As always, we fully support Rockstar’s ambitions and approach.”

This week, Ellie Dunstan, one of the workers fired last year, described the employment tribunal ruling as a “huge moment for us.”

“Rockstar thought they could control the narrative. They’re wrong, and we look forward to proving it. Our case will now be heard in full and put to the test as it should be. The world will get to see for itself the evidence as to what happened last October.

“We loved our work at Rockstar. Losing our passion, our colleagues, and our incomes in the blink of an eye was devastating, and the company management has treated us with disdain ever since.”

Tech companies are often quick to derail unionization where possible, while working conditions in game development have faced particular scrutiny over “crunch,” the practice of employees working extended hours in the run-up to a release. ®

According to Bloomberg, U.S. Commerce Secretary Howard Lutnick has, in a series of recent meetings, told senior ASML executives he’s concerned that one of the Dutch chipmaker’s extreme ultraviolet lithography machines — the EUV systems that are the only tools on Earth capable of printing the most advanced semiconductor patterns — may have ended up in China. That would be a major breach of export controls that have barred ASML from selling EUV to China since the first Trump administration.

It’s a serious claim. Senior administration officials told Bloomberg they have evidence that ASML shipped EUV-related components and transport equipment to China, though they’ve declined, repeatedly, to show it — to Bloomberg or, apparently, to ASML itself. The company says no such machine exists in China and has never existed there. The Commerce Department didn’t respond to Bloomberg’s questions about whether it has evidence of an actual EUV system on Chinese soil.

You might think this isn’t worth paying attention to if you’re outside the chip industry, but it is. ASML is a Dutch company most people have never heard of, but it is, by a wide margin, the most important company in the global AI buildout that isn’t named Nvidia or one of the hyperscalers. It makes the only machines on the planet capable of EUV lithography — the process of printing the microscopic circuit patterns that define the most advanced chips.

Every cutting-edge processor made by TSMC, the foundry behind Nvidia’s and Apple’s chips, depends on ASML tools that took the company roughly two decades and untold billions to develop. There is, at present, no second supplier. That monopoly has made ASML Europe’s most valuable public company, with a market capitalization that has been trading in the neighborhood of $700 billion as of this week, up sharply over the past year on the back of insatiable AI-driven chip demand.

That scale is exactly why the China question matters so much. If even one EUV machine made it into Chinese hands, it would represent one of the most consequential breaches of the export-control regime the U.S. has built over the past several years to keep advanced AI capability out of Beijing’s military and industrial base.

I sat down with ASML CEO Christophe Fouquet six weeks ago, well before this story broke, and asked him directly about the China question.

Fouquet told me ASML tracks every machine it has ever shipped — they’re either in active use with monitored customers or have been dismantled and returned to the company. He said the firm built an internal firewall years ago: employees who can access EUV technology, documentation, and training are walled off from those who can’t, and ASML’s China-based staff sit on the wrong side of that wall by design. He argued the only reason ASML could build an EUV machine at all was that 80% of it already existed from decades of prior knowledge, and that solving the one genuinely new problem — generating EUV light itself — took 20 years on its own. His broader point seemed to be that you can’t reverse-engineer a machine you’ve never had, and nobody in China has had one.

There’s also a simpler commercial logic that cuts against the idea that ASML would risk its export license to quietly arm a Chinese customer. ASML does sell older-generation deep ultraviolet tools to China — gear it first shipped a decade ago — but Fouquet framed that explicitly as a protective calculation, not a loophole. The idea, he suggested, is that it keeps enough of a generational gap that customers can still do business — but without manufacturing its own future competitor. ASML expects roughly 20% of its 2026 revenue to come from already-permitted sales to China. Risking the EUV ban entirely would put that revenue, and the company’s standing as the most valuable monopoly in European industry, on the line over a single illegal sale.

None of this proves the allegations are false. The government hasn’t yet made its evidence public, and it’s worth withholding judgment until it does.

The Commerce Department, under Lutnick’s leadership, agreed late last year to put up to $150 million of taxpayer money into xLight, a startup developing a next-generation light-source technology that’s been written about as a long-term challenge to the core of ASML’s EUV monopoly. xLight’s own CEO told me last year that the company sees itself as a future partner to ASML, not a rival, building hardware meant to plug into ASML’s machines rather than replace them. When I put that framing to Fouquet in May, he was polite about it but unconvinced; ASML, he made clear, doesn’t see itself as needing xLight’s technology to keep its lead.

Does that have anything to do with why Lutnick is suddenly pressing ASML on EUV? Nothing public connects the two. It could be entirely unrelated. But a federal official scrutinizing a monopoly while his own agency has money riding on a startup angling to improve that monopoly’s core technology is worth examining.

xLight isn’t the only outside bet on the future of lithography. Peter Thiel — who has his own long-running ties to Trump’s political orbit — has backed Substrate, a separate startup explicitly pursuing its own EUV-rival technology, with ambitions to compete with ASML more directly than xLight says it intends to.

As Bloomberg notes, a bipartisan bill moving through Congress would go much further than EUV — it calls for an effective ban on all of ASML’s deep ultraviolet (DUV) shipments to China, the less advanced lithography tools that account for roughly a fifth of the company’s expected 2026 revenue. The bill cleared a key committee in April, and the Trump administration hasn’t taken a formal position on it.

Pictured above: ASML CEO Christophe Fouquet

The consultancy giant will take a majority stake in Dragos, and full ownership of RunZero and NetRise.

Accenture is to partially or fully acquire three companies in the area of operational technology (OT) security for critical infrastructure and industrial operations for what it called a “combined enterprise value” of approximately $4.17bn.

The consultancy giant will take a majority stake in Dragos – which Accenture said offers “industry-leading OT threat detection” alongside a “trusted vendor-neutral platform and proprietary dataset” – and full ownership of RunZero and NetRise.

Under the deal, Dragos will continue to function as an independent business while overseeing RunZero, a cybersecurity platform that offers “comprehensive exposure assessment and attack-surface intelligence”, and NetRise, which analyses software supply chains for vulnerabilities.

According to Accenture, combining the three companies, which are based in two different US states, will allow it to advance a platform “to cover the extended environment that controls physical processes” – or ‘xOT’ – at greater scale for the protection of power grids, pipelines, manufacturing operations, distribution facilities and data centres.

“Combining Dragos with RunZero and NetRise will deliver a unified solution that enhances visibility, accelerates threat detection and response, and strengthens Dragos’s ability to scale adoption of its broadened platform,” Accenture said.

Accenture said it expects the three companies to generate, in total, approximately $208m in annual recurring revenue as of June 2026, and noted that its overall cybersecurity business has current revenues of around $10bn, having made a number of OT-focused acquisitions over the past decade.

“Our clients across industries and regions are asking us how to be more proactive and integrated in their approach to cybersecurity,” said Accenture’s CEO and chair Julie Sweet.

Taking on the three companies at a time when “AI-driven cyber threats and geopolitical risk are evolving at a rapid pace … fills this important need”, she added.

Under the deal, which is expected to close in August or September, three executives from the two fully acquired companies will become executives for Dragos, which will continue to be led by its co-founder and CEO Robert M Lee.

“Our energy and water systems, manufacturing plants, data centres and other operational environments need cybersecurity built from the ground up for xOT and designed to keep pace as threats evolve. The consequences of getting it wrong become societal threats,” said Lee.

“Organisations need solutions, not a patchwork of software and services. The addition of RunZero and NetRise will allow the Dragos platform to be a unique, end-to-end platform for global defence, and Accenture will bring its decades of trusted relationships and deep expertise to help us scale and secure more critical infrastructure and physical operations globally.”

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.