Britain's King Charles will host the leaders of major artificial intelligence companies this week to discuss whether a shared set of principles is needed.
Britain's King Charles will host the leaders of major artificial intelligence companies this week to discuss whether a shared set of principles is needed.
Concerns about the water and energy demands of data centers have become prolific in recent weeks. But debates about their impact often overlook the rapid advances in cooling technology and cleaner energy that are already having an impact.
However, data centers’ resource consumption should be front and center of the AI conversion, as it is set to grow rapidly. By 2030, data centers’ water and energy consumption will double as AI demand leads to rapid construction of the infrastructure.
But data centers will not cause the world to run out of water or energy. Water- and energy-saving technologies already in use are reducing their impact, while Shanghai’s groundbreaking underwater data center demonstrates how innovative design can limit water, energy and land use.
Data centers have underpinned the internet since its founding. Apps, email, file storage and cloud services all run through them.
Modern AI workloads, however, have transformed their power demands beyond recognition. Today a single AI data center can match the electricity consumption of 100,000 homes, making them a significant drain on local energy grids, especially if a region is susceptible to shortages.
Data centers’ energy consumption is only rising as more are built across the world. In the US, data centers could increase from 4.4% to 12% of national electricity usage by 2028 — that is equivalent to adding eight New York Cities to the American grid.
Data centers also consume large amounts of water in two main ways. Water is used directly to cool servers and storage systems, which generate considerable heat, and indirectly to produce electricity at fossil fuel power plants where it is heated into steam to drive turbines.
In the UK, data centers have been estimated to use up to eight times as much water as average on hot summer days, underlining how their impact can rise when local resources are already under strain.
As the effects of climate change intensify, the location of data centers, and methods to reduce their water and energy consumption, are becoming ever more important.
Fortunately, resource-saving measures are already being implemented. Data centers on land use closed-loop cooling systems that recycle water rather than continually evaporating it. This can reduce freshwater use by up to 70% compared to traditional open evaporative methods.
China has also pioneered a revolutionary new type of data center in Shanghai: wind-powered, and underwater. The facility, which began full commercial operations in May 2026, uses seawater as a coolant instead of using refrigerated fresh water as land-based data centers do, reducing the electricity required to cool the computers.
It uses at least 30% less electricity than traditional data centers, and offshore wind turbines reduce reliance on fossil fuels and cut the data center’s carbon emissions. Innovative solutions such as these will go a long way to reducing data centers’ resource consumption.
AI is also part of the solution to data centers’ draw on resources. AI can optimise electricity grids and shift computing workloads towards periods when cleaner energy is more efficient, making energy use more effective overall. AI can also anticipate fluctuations in demand and adjust energy consumption accordingly, reducing waste.
It might sound obvious, but some areas are just better equipped to accommodate a data center than others. For example, facilities in hot or water-stressed regions will impose a heavier burden on resources than those supplied by cleaner energy with plentiful water and located in cooler climates.
AI’s growing demand for data centers will increase electricity and water consumption, that is true. But it does not mean the technology will exhaust regional or global supplies of either resource.
The AI boom and the infrastructure buildout accompanying it are well underway. To manage its impact, the industry needs to plan ahead. Cleaner energy, efficient cooling, and sites that have significant water and energy supplies will all limit the infrastructure’s drain on resources.
Earlier this year, an Iran-linked actor used Anthropic’s Claude artificial intelligence model to scrape and analyze data with the aim of helping target U.S. naval forces in the Middle East. A group in Yemen – very likely Iranian-backed Houthi militants – also used Claude to assist with work on guidance systems for ballistic and hypersonic missiles.
These are just some of the revelations about how adversaries and other malign actors have been using one of the premier pieces of AI software being developed in the United States that were contained in a new report Anthropic released late yesterday. The 154-page document also details instances where Claude was used directly and indirectly to support intelligence gathering and surveillance, cyberattacks, influence operations, and even worrisome biological research over the past eight months. In addition to Iran and Yemen, the human actors behind these activities were based in Russia, China, and a host of other countries.
Anthropic has shared details about what it calls the “misuse” of its models in the past, but describes the new report as its “most detailed” to date. The company has clearly been cataloging these case studies, and has assigned each one a control number. Though the entire contents of the report are cause for concern, the explicit attempt to use Claude to target American forces, as well as to develop new higher-end missiles and other military technology, are particular standouts and signs of what’s to come, as we will highlight below.
GTG-30005: Military reconnaissance
GTG-30005 is the case study dealing with the targeting of U.S. naval forces, and the summary is as follows:
“In another investigation, we identified and disrupted an Iran-nexus threat actor that used Claude to collect and analyze publicly accessible data to develop targeting recommendations against US naval forces in the region. The threat actor used Claude to compile targeting handbooks, through a Python pipeline the threat actor built with Claude’s assistance, to identify and track naval positions based on open-source information. The compiled material included a roster of US personnel scraped from captions on public military photographs; publicly accessible ship and aircraft transponder identifiers; commercial satellite-imagery query scripts; and an inventory of public websites that exposed US naval movements. The threat actor also directed Claude to compile vulnerability research on shipboard systems, cataloging known CVEs in maritime VSAT terminals, Cisco communications equipment, and industrial control products.”
“We banned the actor’s account, developed detections to reduce the risk of future misuse, and shared threat intelligence with government authorities to disrupt the threat.”
The Nimitz class aircraft carrier USS George H.W. Bush and several other US Navy warships sail together in the Arabian Sea in June 2026. CENTCOM
How exactly Anthropic defines the term “nexus” here is not immediately clear, and other entries in its new report also use “state-nexus” in addition to country-based variations. However, in this specific case, an addendum is added to the entry that explicitly says that the same account was separately tied to “domestic mass surveillance” software development work “for Iranian state systems,” clearly pointing to an actor directly linked to the regime in Tehran.
The use of AI technology by nation-state armed forces to develop target sets and engage them is already known to be growing globally. The Israel Defense Forces (IDF) have been a pioneer in this regard, but the U.S. military, China’s People’s Liberation Army (PLA), and others are also known to be developing and fielding these capabilities.
GTG-87001: Disrupting a Yemen-based guided weapons engineering cell using Claude to develop guidance software
Anthropic’s new report details an instance where an individual in Yemen used Claude as part of missile guidance work, which is given the control number GTG-87001.
“We identified a cell of threat actors based in northern Yemen running three weapons development programs: a guided rocket that used a commodity phone-class flight computer with final-phase homing guidance; a multi-stage ballistic missile with a stated range goal above 2,000 km; and a multi-variant missile (referred to as the “R2000” set) that included a hypersonic glide vehicle variant.”
“The actors used Claude Code in place of human software engineers to develop the guidance, navigation, and control (GNC) software that steers and stabilizes a flying vehicle. For example, they used Claude to integrate an open-source autopilot onto a phone-class flight computer, writing the control and position estimation software, tuning the control settings, running a firmware build pipeline, and performing a flight simulation. The actors managed several Claude instances at once, assigning each one a role, much as a lead would delegate work on a small engineering team: the actors tasked one instance with writing the code, another with research, and a third with reviewing the code the first instance produced.”
A graphic included in Anthropic’s new report summarizing the ways in which Claude was used by the actors in Yemen to support missile guidance system development. Anthropic
“Our safeguards blocked many of their requests, but not all of them. The actors used a variety of tactics to evade our safeguards, including hiding their goals and the products the software was meant for, and they split their work across multiple sessions so no single session revealed their full intent.”
“These actors carried out a sustained effort to develop guided weapons, including using Claude to design guidance software. We do not have evidence the actors succeeded in fielding an operational device; but they did test-fire a guided rocket. This field test appears to have failed: within hours, the actors returned to Claude to work out why it failed.”
“We identified this activity as part of our internal investigations into suspected weapons development. We banned accounts associated with the actors and shared threat information with public- and private-sector partners to mitigate risks posed by the actors. Nevertheless, we have evidence that the actors had already built an offline simulation toolkit that does not rely on Claude or other engineering computing environments such as MATLAB.”
As mentioned earlier, the report does not name the Houthis, but that group has historically been based primarily in the northern end of Yemen. It is also the only entity in the country known to have active missile development programs, which are conducted with critical support from Iran.
Whether this particular work directly resulted in any new capabilities is unknown. Anthropic notes that there looked to be at least one failed test launch before the accounts were banned, which would have occurred sometime after the beginning of this year, based on the stated cutoff dates for the report. By the end of 2025, the Houthis had already amassed a wide array of very real ballistic missiles, including types with at least claimed ranges of 2,000 kilometers (approximately 1,242 miles). The group also has a variety of cruise missiles and one-way attack drones that it has employed operationally.
An example of just one of the many types of ballistic missiles in the Houthis’ arsenal. Houthi-controlled media
Still, being able to leverage Claude could only have helped any domestic missile development efforts in Yemen, which might help reduce dependency on Iran, at least to a degree. Furthermore, developing functional hypersonic glide vehicles capable of prolonged stable flight has historically been notoriously difficult, even for entities with extensive relevant knowledge bases on an organizational level.
GTG-17001: Disrupting a China-based operation using Claude to draft a fire control specification and acquisition documents for undersea warfare
Use of Claude in foreign military endeavours goes beyond the Middle East, as evidenced by case study GTG-17001. This involved a Chinese actor conducting work related to a naval counter-torpedo defense system.
“We identified a China-based threat actor who used Claude to advance three parallel tracks of work on an anti-torpedo weapons system:”
“First, the actor used Claude to draft a Chinese-language specification for an anti-torpedo fire control system (the core logic that aims and times an anti-torpedo weapon’s response). The document was written to win approval from a Chinese defense manufacturer, which would move the work on to technical certification and operational testing.”
“Second, the actor used Claude to produce a Chinese-language technical proposal of more than 200 pages, accompanied by an executive briefing deck.”
“Third, the actor used Claude to benchmark their own system against specific US anti-torpedo and anti-submarine programs based on publicly accessible information. They then generated a Chinese-language briefing on US Navy systems derived from open-source reporting.”
“The actor presented themselves as an original equipment manufacturer in the US defense sector. We assess the actor was associated with a Chinese defense industry manufacturer aiming to produce a weapons specification and acquisition proposal for the People’s Liberation Army Navy.”
This picture shows a test of an anti-torpedo system fitted experimentally to a Nimitz class aircraft carrier in the 2010s. USN
“The actor used Claude to write the acquisition proposal, refining it over many drafts. After each draft, the actor instructed Claude to role-play a hostile expert reviewer to critique the proposal, then used that feedback to sharpen the next version. In parallel, the actor used Claude to build pieces of the anti-torpedo weapons system’s fire control software and a test matrix to validate them.”
“The operational lift the actor achieved was a function of using Claude to automate complex technical outputs. The actor leveraged the model to compress the development timelines for the certification registry, compliance documentation, and automated fire control logic. The actor also accelerated the traditional human review cycle by having Claude critique the acquisition proposal across multiple rounds of review while role-playing a persona.”
“We uncovered this activity as part of our internal investigations into suspected weapons development. We cannot attribute the activity to a specific entity or actor. But we have banned the account for violating our Supported Regions Policy and our Usage Policy, which prohibits weapons design and development, and incorporated our investigative findings into our safeguards to mitigate the risk of future misuse.”
Anthropic’s report does not name the U.S. programs leveraged here, but does note the information in question was publicly available. Anti-torpedo systems for surface warships and submarines have been an area of significant interest for the U.S. Navy for years now. Those developments have also produced compact torpedo designs for use in other applications.
A Very Lightweight Torpedo (VLT), a compact torpedo design developed by Northrop Grumman. Northrop Grumman
The PLAN’s interest in these same kinds of capabilities is not at all surprising. Underwater threats and countermeasures to them, broadly speaking, would play a central and vital role in any future conflict between the United States and China across the broad expanses of the Pacific.
GTG-27005: Disrupting a Russia-based operation using Claude to engineer an autonomous military drone swarm
Another case study, GTG-27005, details the use of Claude by Russian actors working on drone swarming technology.
“We identified likely freelance Russia-based threat actors who set out to build a full-stack autonomous first-person-view (FPV) kamikaze drone swarm. The actors used Claude Code to write and test the code and save it directly into the actors’ own project files. In addition to Claude Code, the actors used a software-in-the-loop simulation stack and a rented graphics processing host for model training. They called the operation ‘DronDoc’ or ‘Serafim.'”
“The actors used Claude to build the core software system, including the drones’ shared swarm memory and fault-tolerant coordination logic (FTCL); an onboard small language model to govern attack, observe, and return-to-base behaviors; a terminal guidance software system to steer drones to their target (using the onboard camera) and issue the call to detonate; a control-link geolocation module to find opposing drone operators; a passive acoustic detection layer; and low-level logic for the drones’ programmable chips. The actors designed the platform for autonomous lethal engagement; the onboard model could select targets (including a ‘person’ target class) and issue detonation commands without a human in the loop. The actors’ activity—including flashing the low-level firmware to live development boards, provisioning single-board computers, and wiring up a simulation environment over a mesh network—confirmed that they were using real hardware-in-loop testing within their sessions.”
Many were surprised yesterday by the news that a Russian fiber-optic FPV drone flew into Kramatorsk and attacked a car.
But there is nothing surprising here. The war of 2025 is already very different from the war of 2024. From LBZ to Kram — 20 kilometers. Enemy FPVs can fly even… pic.twitter.com/hTfhJFPcxZ
“The actors trained a computer vision classifier on scraped Ukrainian combat footage, splitting the target classes into “enemy” and “friendly,” and allow-listing Russian systems. They also repeatedly used a fixed coordinate in Donetsk Oblast as the demonstration strike point, with front-line cities and corridors in Ukraine as the mission geography.”
“The actors created their accounts between late 2025 and early 2026 and started the operation in mid-May 2026. The actors circumvented our geographic access controls by routing traffic through commercial virtual private servers.”
“We assess the actors were a small, specialized freelance team doing a mix of civilian and military work, not a Russian state entity. We identified nine accounts associated with this group; eight were used only for ordinary freelance work, not weapons-related software development. Based on our investigation, we assess the actors had ties to a regional university with a federal research center associated with the Russian Academy of Sciences. The actors claimed to have received funding from Russia’s Advanced Research Foundation, National Technology Initiative, and Ministry of Defence, though we cannot verify those claims. We identified this activity as part of our internal investigations into suspected weapons development, we banned accounts associated with the actors, and have incorporated our investigative findings into safeguards to reduce the risk of future misuse.”
Anthropic’s report included a table, reproduced below, detailing different types of drones and other capabilities that were fed into the model as part of this work.
GTG-17002: Disrupting a China-based operation using Claude to build targeting software for electronic warfare and air defense suppression
The last case study in the Conventional Weapons section of Anthropic’s new report, GTG-17002, covers electronic warfare-related work by a Chinese actor, with a particular focus on potential air defense targets in Taiwan.
“We identified a China-based actor who used Claude’s chat, coding, and agentic work tools to design, build, and iterate on a Chinese-language suite of about 16 modules for electronic warfare, using the electromagnetic spectrum to detect, jam, or deceive an opponent’s radar and communications, and for suppressing an opponent’s air defenses.”
“The actor used Claude to build the software system, from the underlying logic to the user interface, and iterated through 12 versions. This included implementing and optimizing the system’s radar detection and jamming physics, generating a vulnerability analysis module, and drafting Chinese-language targeting instructions. The software suite the actor built analyzed an opponent’s radars, surface-to-air missile sites, command posts, and communications nodes, then computed their detection coverage, assessed the effectiveness of jamming, ranked targets by value and vulnerability, including which to suppress first, and determined how best to assign jammer sorties to targets across multi-day campaigns. The suite also ranked which of an opponent’s assets to suppress first, and modeled specific engagement envelopes, including those of Patriot and THAAD-class [Terminal High Altitude Area Defense] systems.”
“Mid-project, we observed the actor change the simulation’s default scenario to 12 targets in Taiwan. The targets included a command bunker in Taiwan, an early warning radar site, Patriot and Tien Kung [surface-to-air missile] batteries, major air bases, and a regional combatant command headquarters.”
“The actor also ran a self-hosted model on an internal network alongside Claude and connected the software suite to this model through a tool-use integration.”
“Based on our investigation, we assess the actor is a China-based defense and military-industrial researcher. Account-level metadata and content flagged by our safeguards indicated the actor was linked to PRC research institutions, including the PLA Academy of Military Sciences. We detected this activity as part of our internal investigations into suspected weapons development, we banned accounts linked to the actor, and have incorporated our investigative findings into our safeguards to reduce the risk of future misuse.”
An accompanying table, seen below, mentions several specific types of U.S.-made air defense radars, as well as the Link-16 datalink network, as having come up in the course of this work. Electronic warfare variants of the Chinese Y-8 and Y-9 aircraft, as well as the J-16D electronic warfare jet, and what appears to be a reference to the Golden Eagle drone helicopter, are also mentioned.
A Chinese Y-8GX-11, one of several electronic warfare variants of the Y-8/Y-9 design. Chinese internet via X
The suppression of enemy air defenses is an essential part of any modern air campaign, whether conducted independently or as a part of a larger operation. This would be no less critical in any military intervention against Taiwan from the mainland, or any other larger-scale conflict the PLA might find itself in.
Guardrails and ethics
Beyond the details in any of the specific case studies, Anthropic’s new report underscores growing broader concerns about guardrails for the use of AI and general ethics surrounding the technology. As noted in the instances above, Anthropic has also implemented various safeguards to try to block access to Claude from certain regions and/or to prevent the model from conducting certain types of work. At the same time, what the company has now shared makes clear that those protections can still be circumvented to a significant degree, at least for now.
TWZ has noted in the past that America’s adversaries are likely to be less worried about ethics, in general, when it comes to AI. Anthropic’s new report also shows there are few bounds to what state-aligned or even non-state actors might seek to use this technology for, including the development of potential weapons of mass destruction.
At the same time, AI models like Claude are not going away, and developments continue to advance at an ever-quickening pace, opening the door to further proliferation. Developments are rapidly intermingling globally, with companies in China notably leveraging models created in the U.S. to further their own work, which is then released publicly. The technology will increasingly become more integrated and critical to most aspects of military operations and weapons development, alongside its commercial and civilian applications. It is just as much a superweapon as anything else, and the country with the more capable models, and the willingness to deploy them, will have a major edge in future conflicts.
With all this in mind, it seems very plausible, if not probable, that the case studies described in Anthropic’s new report are just the tip of an approaching iceberg that will need a concerted effort to tackle.
Neurotechnology and artificial intelligence are increasingly converging as neurodata can be collected, combined and reused to develop AI models and applications across different contexts. This creates opportunities for healthcare, rehabilitation and scientific research, but also raises important questions for fundamental rights, democratic accountability and the public interest.
In this context, the EGE calls for a systemic and anticipatory approach to governing neuro-AI. The Statement shifts the focus from individual devices and neurodata privacy to the wider impacts of the infrastructures through which neurodata are collected, processed and reused. It also addresses questions of European technological sovereignty and public-interest infrastructures.
The EGE recommends:
Robust protection of neurodata: define and reinforce protections for neurodata and neurodata-derived inferences under EU data protection law
Responsible development of neuro-AI technologies: ensure possible development of Brain Foundation Models and related neuro-AI infrastructures
Strong Rights protection: ensure that individuals are protected against disproportionate control where neurodata and neurodata-derived data inferences are used in consequential contexts
Enhanced governance capacity: build European public-interest governance capacity for neuro-AI infrastructures
Comprehensive fitness check: conduct a targeted regulatory assessment of the EU framework governing neuro-AI systems
Background
Neuro-AI Infrastructures are interconnected systems through which neurodata are collected, processed and reused to create, deploy and manage AI-powered neurotechnology applications.
It is part of theEthics Advice Mechanism (EthAM), which provides the European Commission (EC) with high quality, independent advice on all aspects of EU legislation and policies, where ethical, societal and fundamental rights issues intersect with the development of science and new technologies.
What do privilege and accountability look like for an AI agent on the wrong side of the law? Evan Ratliff rounds up a collection of challenging cases in which AI agents have functioned as co-conspirators in criminal acts, offered courtroom statements on behalf of deceased victims, and perpetrated deeds that, for humans, are considered to be illegal. Legal teams have already sought to seal their clients’ chats with AI agents, likening the exchanges to conversations with an attorney. Ratliff’s feature is an omnibus, and you’ve likely encountered one or two of his stories via other outlets. Here, the concerns he gathers together reveal the breadth of challenges facing the American legal system.
AI chatbots are already serving as confidants and co-conspirators, assisting and even encouraging humans acting on their darkest impulses. They’re occupying ersatz roles as therapists and lawyers, relationships our legal system typically shields with confidential privilege. They’re producing new kinds of criminal evidence and offering up the beguiling possibility of preventing crime before it happens. And as they’re given more autonomy across our digital and physical systems, they’re even gaining the ability to commit crimes themselves, absent specific human instruction — and, perhaps, in defiance of it.
Singularity is a myth with excellent production values: a clean villain, a clean apocalypse, and a clean date on the calendar. Reality is messier and less cinematic, a slow transfer of wealth, a quiet erosion of privacy, and a data center draining a town’s water supply. For thirty years, the idea that artificial intelligence will […]
AI is playing a bigger role in hiring, but laws protect you. Here is what you need to know.
Europe needs top talent, and employers want high-quality hires – the best person for the job – quickly and in a cost-efficient way. So, given this urgency, it’s no surprise that more and more companies are using AI in their hiring processes. From CV screening to interviews, AI can streamline each stage, reduce hiring times and make the whole process more cost-effective.
But there’s a downside to AI-driven recruitment, as a recent Instagram post highlighted. It describes how a jobseeker from Vienna felt ‘dehumanized’ by an AI interview.
The EU takes AI action
This concern isn’t limited to Austria, which is why the EU AI Actclassifies using AI in employment decisions as ‘high-risk’. Given the growing AI adoption in HR by businesses across the globe (39 % by the end of 2025), the AI Act addresses risks by introducing rules that help make its use more transparent and accountable.
Under the AI Act:
No AI tool can make final hiring decisions without a qualified human in the loop.
Using AI to infer emotions, stress levels or personality traits and assign people numerical scores is banned.
Starting 2 August 2026, companies using AI tools for hiring, evaluating performance or other employment decisions must check regularly for safety risks and bias, keep detailed technical records, ensure human oversight and transparency, and continuously monitor their systems.
Obligations apply not only to the providers of these AI systems, but also the businesses that use them.
These rules even apply to non-EU companies, if their AI system is used or affects people in the EU.
As for the Viennese candidate, here’s what went wrong during the AI interview process:
It repeatedly interrupted the interviewee, jumped to the next question and moved on even when asked to go back.
If the jobseeker paused even briefly during a response, it sometimes took it for a complete reply.
There were errors in the interview transcript.
Instead of answering the applicant’s questions, the AI just recited the company’s ‘About Us’ page.
be informed beforehand on its use and the role it plays in the company’s hiring decision;
ask that a human reviews the hiring decision;
not to have your facial expressions, voice tone or body language analysed by AI during an interview; and
know what data is being collected and, in many cases, to have it deleted once the hiring process is over.
In an increasingly AI-driven labour market, EURES continues to offer human support through its network of trainedadvisers, who can provide information and guidance on practical, legal and administrative matters related to living and working across Europe.
In the meantime, online interviews are here to stay, so gear up for that final step by reading up on some tips for tackling remote interviews.
In this story for MIT Technology Review, Sara Harrison explores the field of social robotics and the growing use of robot companions as therapy for autistic and other neurodivergent kids. These playful machines show promise: Research suggests that they can help children practice social skills and offer something overworked therapists and exhausted parents can’t—a patient companion, available 24/7. But Harrison also reports on their limitations and risks. Does this approach truly work? What about all the data gathered on the child? And what happens when a robot friend suddenly stops working and “dies”? A fascinating and unsettling look at tech that has moved into the most intimate corners of a family’s home.
While Moxie found herself in more and more people’s homes, Embodied still struggled to make money. In 2024, the company announced it would cease operations. Its robots—which depended on external servers that the company could no longer pay for—would descend into a deep slumber.
Videos of bereft children who seemed to have become deeply attached to the blue bot began to circulate online. “I don’t want her to leave,” wailed one child in a TikTok video. Desperate parents posted on TikTok, Instagram, and Reddit looking for solutions. “My autistic child is devastated and I’m pissed,” wrote one parent. “Hope Embodied gives us a couple days to say goodbye,” wrote another.
A senior official within Greece’s National Intelligence Service (EYP), Yiannis Papadopoulos, on Tuesday filed a criminal complaint with the Athens First Instance Court Prosecutor’s Office over the targeting of his phone using Predator spyware. In this, he accuses the four business executives convicted at first instance over the phone surveillance case with a series of […]
“Taste is a new core skill,” OpenAI President Greg Brockman declared earlier this year—one voice in a growing chorus of pronouncements about the centrality of judgment and preference in an economy built on artificial intelligence. And yet taste itself is largely undefined by many of its loudest proponents. Celine Nguyen considers the historic relationship between automation and taste; the “seemingly irrelevant, unimportant” inputs that shape our critical perceptions; and the perils of wielding taste like a cudgel. A striking, insightful blend of criticism and personal essay.
Today’s frontier models are trained on enormous quantities of data. One takeaway from Anthropic’s 2025 settlement with a group of authors, in which the company agreed to pay $1.5 billion (or an estimated $3,000 per book) for scraping copyrighted books from shadow libraries like LibGen, is that its models have “read” at least 482,460 books. But even Claude Opus 4.8 hasn’t written a short story that I enjoy more than those by Lydia Davis or Jessi Jezewska Stevens — both of whom, despite being prodigiously well-read, have almost certainly read fewer books.
This story first appeared as the cover story in Issue 4 of The Atlas, a new global publication by Coda Story and The Continent. Sign up to the newspaper here.
In late July, at the International Congress of Mathematicians (ICM) in Philadelphia, the mathematician Jacob Tsimerman was awarded one of four Fields Medals – the highest honour in mathematics.
A few hours later, Tsimerman announced that he was joining OpenAI.
Tsimerman’s departure from theoretical mathematics, while at the absolute peak of his powers, highlights the fault line in a debate that is upending the normally quiet world of higher mathematics. “What a dick move,” said one mathematics professor, speaking to The Atlas off the record.
Tsimerman, who has written extensively about his fear of “omnicidal futures involving artificial intelligence” – a fear that AI could wipe out humanity, closely associated with the AI “doomer” community – has said he will work on OpenAI’s safety team.
Tsimerman and OpenAI did not respond to a request for comment.
In recent months, OpenAI has announced proofs of various longstanding mathematical problems, solved with one of the company’s latest large language models. Other companies have also made advancements: RedNote, a Chinese AI model, obtained a perfect score at the International Mathematical Olympiad in Shanghai in July.
Human mathematicians, at all levels of the profession, are reeling from the implications. Tim Gowers, a professor at the University of Cambridge who helped to verify one of OpenAI’s proofs, has written that even if AI models fail to progress further, “we have still probably entered an era where it will become very difficult for humans to compete with AI at solving mathematical problems”.
This is a startling assessment – and a marketing coup for the likes of OpenAI.
“Mathematicians actually have a special role in this,” said Joshua Grochow, an associate professor in Computer Science and (by courtesy)Mathematics at the University of Colorado Boulder. “The AI companies’ PR pitch is essentially to say, ‘Look, we have solved math. the intellectually hardest thing ever, therefore we can solve everything.’ ”
LLMs have not solved math – far from it. Nonetheless, Grochow added, recently some capabilities have improved. “They are legitimately able to produce real math, new math that they had not produced before. So you can’t deny that utility. You can deny the extent of that utility, but you can’t say it’s all BS.”
Miek Messerschmidt, a senior lecturer in mathematics at the University of the Witwatersrand in Johannesburg, recently used Anthropic’s Claude to take a crack at a problem that had stumped him. Claude solved it within 30 minutes. “How did I feel? You mean emotionally? I was relieved, because I didn’t have to go down this dead end any more. But I was also disconcerted by what this entity could do in such a short amount of time.”
This year’s Fields medallists: mathematicians Yu Deng (China), John Pardon (US), Jacob Tsimerman (Canada) and Hong Wang (China). Photo: Erin Blewett / AFP.
NEW VARIABLES
Mathematics has been disrupted by technological leaps before. The abacus. The mechanical calculator. The blackboard, even, which played an instrumental role in standardising training and mathematical notation (“It’s hard to research, because most of what’s written on blackboards doesn’t stay there very long,” said mathematics historian Michael Barany).
But AI is a different kind of disruption. Where previous advances helped mathematicians expand the frontiers of their knowledge, large language models threaten the livelihoods, and purpose, of mathematicians themselves.
“Our place in society at developing new ideas has been quite unthreatened for quite some time,” said Tim Chartier, a Professor in the Department of Mathematics and Computer Science at Davidson College. “For a long time, the frontier of mathematics could only be expired by the mathematicians themselves. There were no tools that could push into that boundary, except the research mathematician…At this current time, it’s not like we don’t need research mathematicians – but the landscape has changed so quickly, it’s hard to know what that will look like a year from now, two years from now.”
This matters for everyone: for thousands of years, mathematics has been considered to be the foundation of scientific knowledge. There is no physics without mathematics. No astronomy without mathematics. And there is no computer science – or AI – without mathematics. What happens if there are cracks in the foundation?
“I think that LLMs pose an existential threat to the mathematical community,” wrote Tasmin Chu, a PhD student at the California Institute of Technology, in an essay published on 2 August – the day after OpenAI released a set of ten new mathematical results, resolving or making progress on various problems.
“This is a polemic, not a press release,” Chu continued. “I know that others feel the same way; I have spent today crashing out and learning that top mathematicians I know are crashing out.”
Some of what mathematicians are worried about is the same as what everyone else is worried about: that AI will take their jobs; that AI companies have absorbed everyone’s work, and are reproducing it without attribution or compensation; that students aren’t learning properly any more; and that there won’t be jobs for them when they graduate.
But some worries are more specific to the field. Like the fear that, if mathematics becomes too reliant on particular AI models, then AI companies will have the power to restrict access to a select few – or make it prohibitively expensive. That promising young mathematicians, instead of turning to research, will use their talents instead to refine proprietary algorithms, which may or may not be used for the common good. That there will be no one left to understand, let alone vouch for, the mathematics produced by the machines. That objective truth, or anything close to it, is going to be much harder to prove – a fear shared with journalists, incidentally, who find it increasingly difficult to verify whether a photograph is real or fake.
2nd Century relief of a merchant using an abacus. Photo: Musée de La Cour d’Or - Metz Métropole.
MATHEMATICIAN IN THE LOOP
After speaking with more than a dozen leading mathematicians from around the world, The Atlas can confirm that Chu is not alone with her fears: it’s all anyone is talking about. The crisis is existential – and mathematicians are looking to one man in particular to guide them through it.
There are no working mathematicians who are truly famous. The closest anyone gets is Terence Tao, a professor at UCLA, who is often referred to as the world’s greatest living mathematician. A prodigy, Tao skipped five grades of school, receiving his masters degree at age 16. By 21, he had earned a PhD from Princeton, and was awarded the Fields Medal just ten years later, in 2006, for his pioneering work with Ben Green on prime number theory.
But what sets Tao apart, in a field infamously incomprehensible to outsiders, is his ability to communicate to both mathematicians and laypeople alike. This has made him a de facto spokesperson for the mathematics community – and extremely influential in setting its norms. “From all accounts he’s a lovely man,” said one researcher, speaking anonymously.
“But the social pressure to agree with Terry Tao is really strong.”
On July 24, the day after Jacob Tsimerman joined OpenAI, Tao gave a public lecture at the ICM titled “Mathematics in the Age of AI”. In it, he presents his “working hypothesis”: that AI tools for mathematics, which he already uses in his own research, will keep getting better. This will allow for collaboration on a vastly greater scale than has ever been possible before, Tao argues, and speed up the pace of research – revealing new, previously unimagined frontiers.
But it will also change, fundamentally, the role of human mathematicians. There will be little point in problem-solving – the computers will handle that. But the humans will still have to do the really hard work: understanding why AI-generated proofs are correct, and explaining this to others. The sooner they embrace this and plan accordingly, Tao concludes, the better.
The Atlas reached out to Terence Tao to request an interview. His human assistant, UCLA media relations officer Holly Ober, declined the request, making a referral instead to a “living document” which is maintained by Tao’s Claude AI assistant, on his website. This document synthesises Tao’s current thinking on AI, based on his public writings and talks, even “interviewing” Tao with follow-up questions.
As per his AI assistant, Tao believes that the topic of AI in mathematics is too important to be left to just a small number of spokespeople – and that his own “influence on this discussion should diminish as the rest of the community steps up.”
Terence Tao, often referred to as the world’s greatest living mathematician. Photo: OpenAI Academy.
AN UNSOLVED EQUATION
Not everyone believes in the inevitability of AI defining the future of mathematics. According to several sources, both organisers and funders were disappointed with the turnout at the ICM in Philadelphia. Despite the gravity of the occasion, the congress was not at capacity.
Ila Varma and Tarik Aougab, associate professors at the University of Toronto and Haverford College respectively, may have had something to do with that. In March, they launched a petition to boycott the ICM due to the ongoing war in Iran, increasing authoritarianism in the United States and the risks faced by visiting academics – if they were even allowed in. “AI companies were invited, but most African mathematicians who tried to attend couldn’t get visas,” Varma said.
More than 2,600 people signed the petition. For Varma, this is evidence that mathematicians may yet use their status to resist the influence – and the cash – of the big AI companies.
“There’s an idea that the positionality of mathematics in society is to decide who’s smart, who’s worthy of the most lucrative career opportunities and the privileges that go along with it. And now we are being asked to take this technology that is likely to be extremely unhealthy for users, that’s bad for the environment, this thing built on unethical labour, and our discipline is being used against many of our wishes to tell society that this thing is intelligent, and that you should let it do your thinking for you.”
Michael Harris, a professor at the Institut de Mathématiques de Jussieu in Paris, was at the ICM in Philadelphia. “The talks I attended on AI, starting with Tao’s public lecture, took pains to insist on the need for human mathematicians to make sense of anything produced in the name of the discipline, by machines or by humans,” he said. “There is a growing resistance to any use of AI whatsoever in mathematics, especially among young people, and this will become evident over the next few months.”
One of the clearest voices of dissent belongs to Max Weinreich, Assistant Professor of Mathematics at the City University of New York.
In an essay published earlier this month, Weinreich makes the case against what he describes as “artificial mathematics”: “Mathematicians do not need to use AI, nor should we use AI; soon, I argue, you will not want to have your name associated with AI. Instead, we should use our intellectual authority to oppose its development.”
His opposition is about more than mathematics, he told The Atlas. In his view, the environmental and financial implications of the AI boom pose major societal risks – and mathematicians, given their privileged position, are one of the few groups of people who can do anything about it.
As John Fletcher, a theoretical physicist and mathematician, observed: “There is a lot of power in being perceived as the smartest people in the world.”
Po-Shen Loh, a professor at Carnegie Mellon and former coach of the US International Mathematical Olympiad team, backs mathematicians to figure it out – by treating it like a maths problem. “AI is making mathematicians engage with the real world. They are smart enough to realise they have an existential crisis. They’re quickly racing down to discover all the dead ends, to see what works and what doesn’t.”
And when they figure it out, maybe they’ll let the rest of us know.
A version of this story, edited for length, first appeared in The Atlas newspaper. Additional reporting by: Shougat Dasgupta, Masho Lomashvilli, Irina Matchavariani.
Kazakhstan is moving rapidly to position artificial intelligence at the center of its economic modernization strategy, investing in computing infrastructure, skills, startups and Kazakh-language technologies as governments worldwide compete for a place in the emerging AI economy, writes Assel Satubaldina.
Kazakhstan has an increasingly ambitious AI agenda. The government has made AI part of its broader economic modernization strategy, betting that it can improve public services, attract investment and help the country generate more value beyond its traditional reliance on raw materials. It has been investing heavily in computing infrastructure, training, startup ecosystems, and Kazakh-language models. At the 38th plenary session of the Foreign Investors’ Council in July, President Kassym-Jomart Tokayev explicitly called the integration of AI into state investment policy a “critically important priority”.
The government has adopted a Digital Code and a law on artificial intelligence, launched the Digital Qazaqstan strategy, established a Ministry of Artificial Intelligence and Digital Development, and designated 2026 as the Year of Artificial Intelligence and Digital Development. This push comes at a time as AI is becoming both a major global industry and an increasingly important source of economic and strategic power. UN Trade and Development estimates the global AI market will rise from $189 billion in 2023 to $4.8 trillion by 2033. This is a 25-fold increase in just a decade.
Globally, AI push is concentrated among a handful of countries. U.S. private AI investment alone reached nearly $286 billion in 2025, more than 23 times the level in China, according to Stanford University’s 2026 AI Index. Global corporate AI investment more than doubled in 2025, while private companies produced more than 90% of the year’s notable AI models. For emerging economies, the concentration of AI infrastructure in a small number of countries can make this race appear expensive, not to mention that training large language models requires massive quantities of computing power, electricity, connectivity and capital. Kazakhstan is unlikely to compete with the United States or China at the most capital-intensive layer of the AI industry nor does it need to. Experts say its opportunity lies in combining global AI technologies with domestic expertise, while building its own capabilities in areas where dependence creates economic or strategic vulnerabilities.
Strong foundation The country’s AI push builds on years of its experience of deploying digital technologies at a scale. Advanced technologies are increasingly embedded in government services and banking. “Right now, especially in the artificial intelligence applications in fintech, Kazakhstan is one of the most progressive countries in the world,” Huseyin Atakan Varol, dean of the School of Computing and Artificial Intelligence and director of the Institute of Smart Systems and Artificial Intelligence at Nazarbayev University, told The Astana Times. He pointed to technologies including palm-recognition authentication and AI-powered banking assistants as examples of how quickly digital tools have entered everyday life. International indicators tell a similar story but also expose an important weakness.
Kazakhstan ranks 10th globally for online government services, according to figures from the World Intellectual Property Organization. Yet it ranks 115th for university-industry collaboration and 112th for venture-capital deal activity relative to the size of its economy. Dr. Varol said Kazakhstan still allocates a relatively small share of its economy to research and development, with the government providing most of that funding. While spending on science has increased in recent years, spending on research and development does not reach 1%. “Nearly all of this allocation comes from the state, very little from the private companies,” he said. “So basically, industry, academia, government interaction, the triple helix structure is not yet fully working in Kazakhstan.”
That may prove to be one of the important constraints on Kazakhstan’s AI ambitions. While the government can establish institutions, purchase computing infrastructure and train specialistics, building a durable innovation economy requires more. Alem.AI center in Astana is intended to help close that gap by connecting education, research, startups and companies. “We have some real and simple goals. The first one is to build a community of professionals, who not just use but understand what AI actually is. It means that we train different categories,” Alyona Slavkina, head of the Alem.AI Development Office, told The Astana Times.
The community is supposed to include schoolchildren, students, working professionals, researchers and entrepreneurs. The idea is to create a pathway from first exposure to AI to education, research and eventually the development of commercial products. The challenge, however, is that AI is evolving faster than conventional education systems. Olzhas Urazalin, head of Tomorrow School, said technologies and development tools can change within months, while formal academic programs often require years to adapt. “Many graduates have strong theoretical knowledge, but very limited experience working with modern AI tools, real data sets, and product development.
\This indeed creates a gap between learning and creation. Young specialists often learn how to use existing AI solutions, but they rarely get the opportunity to understand how these systems are built, or how to develop new ones,” Urazalin said. Tomorrow School is built around an alternative model. Students learn primarily through projects rather than conventional lectures and work on real problems brought into the curriculum by businesses. The program is free, runs for 18 months and is open to applicants without traditional technology backgrounds. “In my opinion, Kazakhstan must focus on developing, first of all, creators, not only consumers of AI technology,” he said, highlighting talent, infrastructure, and international collaboration as priorities.
Development pathways Dr. Varol describes two broad pathways for technological development. One is indigenous development, which requires substantial investment, time and a long-term commitment. The other is what he calls “technological piggybacking,” adopting technologies already developed elsewhere. For Kazakhstan, Varol said that generative AI requires elements of both approaches. “We have been also debating and discussing a lot about what is the best strategy for Kazakhstan, especially in the development of generative AI technologies. Our internal deliberations always pointed out that we must develop these technologies indigenously as well,” he said.
Varol also stressed that developing domestic capabilities is not only about owning a model, but also about creating a pool of specialists with expertise who can later move into startups, established companies and new commercial products. Recent events also demonstrated that access to foreign technologies can become subject to government restrictions. In June, the US government temporarily imposed export controls barring foreign nationals from accessing Anthropic’s Fable 5 and Mythos 5 models, including foreign-national employees of Anthropic itself. “Creators of these technologies can stop your access to these technologies and this creates a very big dependence risk,” said Dr. Varol.
Technological sovereignty, however, should not be confused with technological self-sufficiency. Mohamed Eissa, chief investment officer at the International Finance Corporation, emphasized that this explains why emerging markets are unlikely to compete directly at that layer of the industry. “You are not seeing LLMs [large language models] being developed in emerging markets because you need compute, connectivity, power,” Eissa said. “Those things are big infrastructure gaps in emerging markets.” But he draws an important distinction between building AI infrastructure and using AI.
He pointed to a rapidly growing AI company in content creation that originated from Kazakhstan as an example. “But using AI in emerging markets, you don’t need to build that to be able to use it. That is an opportunity in emerging markets: finding innovative people who have really good ICT skills, have focused on a business problem and now have the tools to compete with Silicon Valley,” he said during The Development Podcast on the World Bank Group YouTube channel.
That may be one of the most important opportunities for Kazakhstan. According to him, in emerging markets, the much larger opportunity may be the application of AI to businesses that already exist. Likewise, in Kazakhstan, potential economic impact of AI should not be measured only by how many programmers, AI startups or research laboratories the nation generates, but rather by whether this technology raises productivity gains across the wider economy.
Existing strengths Dr. Varol said some of the applications of AI lie in sectors where the country already has an expertise. Mining, energy and logistics are among them. “Kazakhstan is one of the richest countries with respect to mining and natural resources, but most of these natural resources are exported with a small amount of added value to other countries,” he said. Using AI and digital technologies in those industries, he noted, could allow Kazakhstan to generate more value from resources before they leave the country. Where dependence matters There are areas, however, where Kazakhstan has stronger reasons to maintain domestic capabilities. Dr. Varol said questions about where information is processed and stored become significantly more consequential when AI systems are used by companies or public institutions.
“In general, I believe this is also very important with respect to the sovereignty aspects of a country,” he added. Language presents a different form of dependence. Global AI systems are trained on enormous quantities of data, but languages are represented very unevenly in those datasets. Dr. Varol said approximately half of publicly accessible internet content is in English, compared with only around 0.03% in Kazakh. That means developing Kazakh-language AI starts long before a model is trained. ISSAI, which was founded in 2019, has developed the country’s first LLM in 2024. Drawing on the institute’s experience, Dr. Varol explained that researchers first need to collect texts and other data, remove poor-quality material, organize the information and convert it into formats suitable for machine learning. Researchers faced another problem once a model has been trained on how to determine whether it actually understands the language. “Because the internationally accepted method for doing this is called model benchmarking and all model benchmarks are in English language or Chinese language, but none were in Kazakh language. Basically, we also have created a suite of benchmarking data sets,” he said.
The development of Kazakhstan’s LLM illustrates another benefit of indigenous AI development. The process itself creates capabilities. The team had to recruit and train specialists while learning to operate multi-node AI computing infrastructure, expertise that had not previously been used at that scale. The significance of the project therefore extends beyond the model it produced. It helped build a domestic pool of engineers with experience in the infrastructure and workflows required for advanced AI development. Kazakhstan’s AI strategy therefore does not require choosing between complete technological independence and reliance on global platforms. The more realistic task is to determine which capabilities the country needs to control, which technologies it can efficiently import and where AI can create the greatest economic value.
Nvidia on Wednesday forecast a 70 per cent jump in revenue next fiscal year, underscoring unabated demand for AI computing, while warning about shortages of memory components.
Login
