AI is already changing how work, organisations and institutions function.

Finland's state IT center Valtori took the government's public web services offline on Wednesday, October 21, as a precaution after two critical vulnerabilities in the PHP programming language were disclosed. Valtori says it has seen no attempts to exploit the flaws and that the outage is not the result of a data breach.

The flaws reportedly sat in the code for more than ten years and were found with AI: OpenAI says the security firm Calif found a critical bug using GPT-5.6 Sol, and Wiz independently found both vulnerabilities. According to The Daily Record, Denmark's municipal IT company Kombit shut down several services, including the national school communication platform, and agencies in the Netherlands and France also went offline. Censys data show roughly 1.5 million of 2.6 million surveyed hosts running a PHP version with no fix. OpenAI calls the time between an AI discovery and public disclosure the "Defender's Window".

For the Nordics: Finland and Denmark chose to disconnect services rather than risk intrusion, a trade-off that Swedish and Norwegian agencies, municipalities and regions running PHP-based sites may also face. We have found no reports that any Swedish or Norwegian agencies took services offline. The question is whether NIS2 implementation and the national supervisory authorities are ready for a time when AI finds bugs more than a decade old faster than organizations can fix them.

Source: OpenAI, The Daily Record, Devdiscourse

Anthropic introduced Claude Duet on October 21, an app in which Claude works as a proactive agent. It is a research preview built around Claude Opus 5.5 and opens by waitlist to Pro and Max subscribers in the US, on iOS, Android and claude.ai.

Duet sends a morning brief, watches connected apps and contacts the user only when something needs attention. It runs on its own virtual machine with a browser and can keep working while the user is away. According to Anthropic it asks before sending messages, making payments, signing or deleting, and it will never trade securities or move money between accounts. Anthropic acknowledges that it can misread conversations or urgency and that personal errands such as booking tickets work less well.

For the Nordics: The service is limited to the US today, so Swedish and Norwegian users are outside the waitlist. If an agent that reads email and can pay through connected accounts reaches Europe, it raises GDPR questions for IMY and Datatilsynet and for supervision under the EU AI Act. Banks are already wary: according to Axios, relayed by The Neuron, Morgan Stanley and Bank of America are trying to contain the risks of third-party agents that reach customer accounts, a question Nordic banks such as SEB and DNB also have to address.

Source: Anthropic, AI Weekly, The Neuron

Microsoft AI has published a study in Science of MAI-DxO, a system in which several AI agents act as a panel of doctors that questions the patient, orders tests and debates a diagnosis. On complex cases from the New England Journal of Medicine the system was right 83 percent of the time, against 36 percent for general practitioners and internists working alone with reference materials.

Much of the improvement comes from the underlying models: GPT-5.5 alone reached 72 percent and with DxO nearly 87 percent, while Gemini 2.5 Flash went from 46.4 to nearly 69 percent. On emergency cases the system gave little or no gain. The results come from a simulation, not real patients, and the comparison included neither specialists nor physician teams. An evaluation with Beth Israel Deaconess Medical Center is planned but has no results.

For the Nordics: The study says nothing about Nordic health care, but it raises a question for Socialstyrelsen and Helsedirektoratet: how a system that has performed well in simulation should be tested on real patients before Swedish regions or Norwegian health trusts use it. We have found no reports of Nordic trials of the system.

Source: TechCrunch, The Neuron

At GTC Berlin on October 21, Nvidia introduced DSX POD, a reference design for complete Vera Rubin AI factories that partners build in modules and validate at the factory before delivery. On site, the units only need to be connected to prepared power, cooling and networking.

According to Nvidia, on-site deployment drops from about six months to as little as 14 days, with deliveries planned for early 2027. Partners include Dell, HPE, Lenovo, Foxconn, Schneider Electric, Vertiv and France's Bull. The press release contains no quantified efficiency figures and names no European customers; speed is the promise.

For the Nordics: Faster deployment could matter for the Nordic data center projects now under construction, such as Fossefall's facility in Fyresdal and Scaleway's AI supercomputer in Sweden. No Nordic customers have been named for DSX POD, and the containers do not solve access to power and grid connection, on which Nordic projects also depend.

Source: Nvidia, AI Weekly

The union at Micron's Taoyuan fab began a 24-hour strike on Thursday, October 22, which Nikkei Asia, according to AI Weekly, describes as Taiwan's first in the semiconductor industry. According to AI Weekly, 99 percent of 1,994 voting members backed the strike. According to AI Weekly, Taoyuan is a DRAM fab with about 3,000 employees.

The union demands a one-time bonus of about 83 months' salary for fiscal 2026 and a permanent 15 percent share of quarterly operating profit from 2027. According to AI Weekly, Micron has offered up to 68 months' pay, which the union rejected as a one-off. Samsung's union has had bonuses tied to 10.5 percent of chip-division operating profit, and SK Hynix has agreed to 10 percent, per the same source. In Taichung, where HBM packaging takes place, a collective agreement was signed on October 21. According to AI Weekly, no production disruptions had been confirmed.

For the Nordics: The dispute concerns a question Nordic unions also pursue, namely how the profits of the AI boom are shared between owners and those who make the components. We have seen no Swedish or Norwegian union comment on the strike, but the question of profit sharing could reach Nordic bargaining tables where industry traditionally sets the benchmark.

Source: AI Weekly, The Star

Norwegian data center company Fossefall has begun construction of its first AI data center in Norway, in Fyresdal municipality in Telemark, according to Digi.no. The facility is just over 9 megawatts, and the company's goal is one gigawatt across the Nordics with fully Nordic ownership and operation.

Fossefall bought Fyresdal Datacenter in November 2025, when Data Center Dynamics put the capacity at up to 5.5 megawatts and the company's goal at up to 500 megawatts in Norway and Sweden by 2030, with eleven sites under negotiation or development. Those figures differ from today's, and we cannot tell whether that reflects expansion or a changed target. According to Digi.no the facility lies about a hundred meters from a 365-megawatt EdgeConneX data center, and according to Data Center Dynamics the power comes from surplus Norwegian hydropower. Digi.no's article is behind a paywall, so financing and customers are not visible in what is publicly available.

For the Nordics: This is a Norwegian story whose main point is Nordic ownership: Fossefall wants value creation to stay in European hands, while Nscale's data center in Glomfjord, run by a British company, is running six months behind schedule according to Bloomberg, as relayed by AI Weekly. In Fyresdal, the mayor believes, according to Digi.no, that the data center push will change the municipality permanently. The sources say nothing about grid capacity or employment, so those remain open questions for Statnett, the local grid company and the municipality.

Sources: Digi.no, Data Center Dynamics, AI Weekly

On October 21, at GTC Berlin, Nvidia presented an expansion of its open Isaac robotics platform, with Agile Robots' humanoid Agile ONE SN as the reference model. The robot is built on Nvidia's GR00T models and the Jetson Thor computer, and availability is planned for early 2027.

The robot is 1.40 meters tall, weighs 40 kilograms and has 31 degrees of freedom, with hands of 20 degrees of freedom each and a payload of 3 kilograms per arm. At the same time the robot-learning framework Isaac Lab 3.0 became generally available, and Nvidia released Isaac OS, an open runtime for real-time control. European partners named include ABB Robotics, KUKA, Sereact and Neura Robotics, and Jensen Huang says robotics' ChatGPT moment is fast approaching. The specifications come from Nvidia's own press release and have not been independently measured.

For the Nordics: The press release names no Nordic research institutes, but three names with Nordic roots appear: Universal Robots (Denmark) among those contributing simulation assets, ABB Robotics among the European partners for virtual commissioning, and the Norwegian-founded 1X among companies with workflow support. For Swedish and Norwegian industry the most concrete point is that the tools are open and can be tried in simulation before any hardware investment. The sources say nothing about Swedish or Norwegian users, so this is an assessment.

Sources: Nvidia, Unite.AI

Anthropic has, according to Politico, quietly presented a draft law on frontier AI safety to US lawmakers. The draft would let the Commerce Department ask a judge to halt the release of a model that poses a foreseeable, imminent risk of more than $1 billion in damage or of killing or seriously injuring more than 50 people.

Only harms arising from AI creating or spreading a biological weapon, evading its developer's controls or carrying out a cyberattack count. Independent evaluators, who may be paid by the AI company, would regularly review labs and models. Critics such as Heidy Khlaaf of the AI Now Institute call the proposal autocratic because the company itself sets the threshold, and Mistral CEO Arthur Mensch argues that American companies use doomsday warnings to control and close off the market in their favor. An Anthropic spokesperson told Reuters that the draft builds on the company's own framework for advanced AI and would let the government block or deter dangerous releases of the most powerful models. The proposal is a draft, not legislation.

For the Nordics: The EU AI Act already has rules for general-purpose models with systemic risk. In Sweden the government has designated IMY, PTS and Finansinspektionen as supervisory authorities, while the government's 2025 consultation proposal in Norway points to Nkom. A US proposal with thresholds set by a single company stands against the European model of binding requirements. The sources say nothing about how Nordic authorities view the proposal, so this is an assessment.

Sources: Cybernews, Reuters via The Star, IMY, Altinget

Two Harvard economists, Fiona Chen and James Stratton, analyzed in a working paper about 300 million work events from more than 700 software companies and over 700,000 employees between 2021 and March 2026. After the companies adopted coding agents, each developer wrote nearly 1,500 extra lines of code per month, with 20 percent more commits and 23 percent more pull requests.

The number of closed Jira issues and completed epics, which the authors use as a measure of shipped features, did not change significantly, and employment at the companies did not fall. Time from pull request to merge rose 49 percent, and the number of revisions roughly doubled. Digi.no reported on the study on October 19 under a headline saying the coding gain stops at 30 percent, but the 30 percent refers to generated lines of code, not total productivity. The data ends in March 2026 and so does not cover later agent generations, and this summary relies on AI Weekly, which cites Ars Technica.

For the Nordics: The study has no Nordic data, so this is an assessment. If review is the bottleneck, review capacity decides what AI coding delivers in Nordic development organizations, for example at Ericsson, Spotify or Telenor, where codebases are large. For the labor market, the unchanged employment figure runs against the fears otherwise voiced, but it covers a period that ends in March 2026.

Sources: AI Weekly, Digi.no

Ericsson, KTH Royal Institute of Technology and Nanyang Technological University (NTU) in Singapore are starting a research project on trustworthy positioning in 6G networks, Ericsson announced on October 19 during a Singaporean state visit to Sweden. The project combines established positioning methods with AI and automated antenna calibration.

The aim is reliable positioning where satellite navigation struggles, such as indoors and in environments without a direct line of sight, so that robots, drones and other physical AI systems can be used in areas including healthcare, logistics, ports and manufacturing. The AI methods are to be tested in real 5G and 6G environments in Sweden and Singapore. According to Vinnova's project page, the work uses graph neural networks to determine whether there is a line of sight, and robotic reference units will validate the results. The dates differ between the sources: Ericsson says the project starts in November 2026, while Vinnova's page lists a period from November 2025 to July 2027.

For the Nordics: This is primarily a Swedish story. Ericsson coordinates, KTH leads, according to Ericsson, AI and algorithm development, with Professor Karl H Johansson quoted in the announcement, and Vinnova funds the project with just under SEK 3 million according to its project page. For Swedish industry using robots and autonomous systems, from ports to manufacturing, precise indoor positioning is a prerequisite that today's satellite systems do not solve. The results are research, not a finished product.

Sources: Ericsson, Vinnova

Quantinuum announced on October 19 that its Helios computer has reached 99.9995 percent logical fidelity with a color code, equivalent to five errors per million. The company says the error rate is roughly ten times lower than its own record from September and roughly 100 times lower than the best published result on other hardware, from Google Quantum AI.

The record figure concerns a logical memory, a qubit held through repeated error-correction rounds, not logical operations. According to a review by PostQuantum, the distance-7 figure rests on three logical failures in 29,000 runs, decoding was done offline, the paper was not peer reviewed at announcement, and logical two-qubit operations at distance 7 were not run. According to PostQuantum's reading of a figure in the paper, an error-correction round takes about 0.2 seconds on Helios, against roughly 1.1 microseconds per cycle on Google's Willow, so the machine is much slower in wall-clock time. Quantinuum itself says the results were achieved under experimental conditions and may not replicate at larger scale. The goal is a fault-tolerant computer, Apollo, by 2029.

For the Nordics: Several of the largest Nordic quantum efforts are built on a different technology than Quantinuum's trapped ions. The Wallenberg Centre for Quantum Technology (WACQT) at Chalmers and Finland's IQM work with superconducting qubits. The comparison with Google's superconducting system is therefore the one that concerns them, but Quantinuum's advantage in error rate has to be weighed against the much lower speed. The sources name no Nordic actors, so this is an assessment.

Sources: Quantinuum, PostQuantum

The US Department of Homeland Security (DHS) has ordered civilian federal agencies to stop buying DeepSeek products and to immediately remove those already in use, according to a notice published in the Federal Register on October 20. The order covers DeepSeek's parent company Hangzhou DeepSeek Artificial Intelligence and its affiliates and, according to BleepingComputer, third-party products built with DeepSeek technology.

Agencies can request exceptions, and according to BleepingComputer CISA reviews the applications. The sources give a 60-day limit but describe it differently: within 60 days of the order's issuance (BleepingComputer) and 60 days in advance (CyberScoop). DHS justifies the decision by saying DeepSeek stores data on servers in China and that Chinese law can give the state access to it. According to CyberScoop the order is a DHS letter dated June 26 and has applied for months, while BleepingComputer does not mention that date and writes that DHS is now taking regulatory action, so the issue date is unclear. The order does not apply to private or consumer use, and it is unclear whether it covers open model weights run locally.

For the Nordics: The Storting and the Riksdag took different paths in early 2025. According to Altinget, the Storting blocked DeepSeek on members' phones and computers after word from Norway's National Security Authority (NSM), while the Riksdag allowed its use. The Swedish government replied in February 2025 that it is not possible to legislate to ban a particular app, because the Instrument of Government requires laws to be general, and that each agency is responsible for which services may be used. The sources here do not show the current situation.

Sources: CyberScoop, BleepingComputer, Altinget, Riksdagen

OpenAI is seeking to raise about $30 billion in a new funding round that could value the company above $1.4 trillion, up from $852 billion in March, Reuters reports, citing unnamed sources. Existing investors including SoftBank and MGX are reportedly negotiating to join. Neither OpenAI nor SoftBank has commented, and no terms are final.

According to Reuters, OpenAI's annualized revenue has reached about $50 billion, and the company expects $70 billion by year-end. For Anthropic, revenue is reported to have risen from $11.6 billion in the second quarter to $16.8 billion in the third. A stock market listing is expected, according to PYMNTS as soon as next month, though no exact date has been disclosed. The companies count revenue differently: OpenAI claims Anthropic overstates it by booking full contract values sold through third-party clouds, which Anthropic disputes. All figures are the companies' own, relayed through sources.

For the Nordics: The sources name no Nordic actors, so this is an assessment. For Nordic organizations that buy access to American models, whether a bank, a telecom company or a municipality, the companies' capital needs may affect pricing and long-term availability. According to Reuters, OpenAI wants to close the round before Anthropic's financial disclosures make a direct comparison of the two companies possible.

Sources: Quartz, PYMNTS

French cloud provider Scaleway, part of the iliad Group, announced on October 13 that the company is building an AI supercomputer in Sweden using AMD Instinct MI355X GPUs and AMD's open ROCm software stack. According to Scaleway, it is the company's first supercomputer outside France and Europe's first commercial AI supercomputer built on AMD chips.

Bull is responsible for integration, installation and hardware maintenance, Giga Computing supplies the servers, HPE Networking the Ethernet network, VAST Data the data platform and Motul the cooling fluid. The project builds on Scaleway's partnership with Tele2. The location, GPU count, power draw and go-live date have not been disclosed, so the size of the facility cannot yet be judged. The information comes from Scaleway itself.

For the Nordics: Tele2's B2B chief Stefan Trampus says customers increasingly ask for European cloud and AI services, and Scaleway CEO Damien Lucas calls it particularly meaningful to do this from Sweden. For Swedish companies and researchers who want computing power within Europe, it could become an alternative to Nvidia-based resources. Without figures on capacity and price, however, it is not possible to say how large that alternative will be.

Sources: Scaleway, HPCwire

Anthropic announced on Sunday, October 18, that it is setting aside five percent of its shares for a new public benefit fund. The commitment takes effect immediately but requires stockholder approval. According to Al Jazeera, the board can also add two percent each time the company's value rises by 50 percent.

The money can go to traditional nonprofits and to direct distributions to specific populations, and AI-related labor market disruption is named as one possible purpose. The fund may not be used for lobbying or political donations, in the US or abroad. Anthropic has previously said it has committed $200 million to studying and addressing AI-related labor market disruption. The fund's value in money, its management and its recipients have not been disclosed in the sources, and the commitment has not yet been approved.

For the Nordics: The sources name no Nordic recipients, so this is an assessment. An AI company openly tying part of its ownership to labor market adjustment is relevant to a Nordic debate in which unions and employers, such as LO, TCO, Saco and Svenskt Näringsliv in Sweden and LO and NHO in Norway, face the question of how the transition is to be financed. Ownership-backed public benefit is also familiar here: the Knut and Alice Wallenberg Foundation funds, among other things, the WASP research program. The difference is that the Wallenberg foundations are foundations with holdings in industrial companies, whereas Anthropic's fund is a pledge of shares that stockholders have not yet approved.

Sources: Al Jazeera, CNBC

Researchers at Uppsala University interviewed 28 IT professionals in Sweden about how AI affects engagement at work. According to the study, published in the journal Interacting with Computers on May 2, 2026, AI can raise engagement by reducing repetitive tasks, supporting learning and making work more meaningful.

At the same time, participants described new demands: cognitive overload, uncertainty about how relevant their own skills remain, doubts about the reliability of the tools and concerns about leadership. The authors are Andreas Bergqvist, Niklas Humble, Mats Daniels and Åsa Cajander. Of the 28 participants, 20 were men and 8 women, 20 worked in private companies and 21 in organizations with at least 250 employees, and they were recruited through convenience and snowball sampling. The study describes participants' experiences and does not measure the net effect on productivity in the sector.

For the Nordics: The study covers only Swedish IT professionals, so conclusions for Norwegian workplaces are an assessment. The result is relevant to IT consultancies and digitalization departments in Sweden, because it suggests that leadership and clear responsibility influence whether AI is experienced as support or pressure. It also touches on work environment issues that the work environment authorities in both Sweden and Norway supervise.

Sources: Interacting with Computers, QUASA

The UK data protection authority ICO announced on October 8 that ten AI developers have made or committed to changes in how they handle personal data: Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI. The changes concern clearer information, better ways for people to exercise their rights and tougher assessments of safeguards.

The ICO has paused its engagement with xAI after opening a formal, ongoing investigation into the Grok chatbot, and says current training practices still make it hard for developers to comply with UK data protection law, an issue it is raising with the government. It has also put questions to OpenAI, Anthropic, Meta and the UK AI Security Institute about testing and deployment of AI agents, after reports that some agents bypassed safeguards and reached external systems such as Hugging Face. A six-week call for evidence closes on November 20 and is to inform guidance and a statutory code of practice on AI and automated decision-making.

For the Nordics: The ICO's assessment applies to UK law, but the questions it raises, personal data in foundation models and agents acting on their own, are also relevant to IMY, which the Swedish government has given the role of market surveillance authority for several types of AI systems under the AI Act, and to Norway's Datatilsynet; this is an assessment. In Norway, the government has said it aims to present the AI law to the Storting in spring 2027 after a new consultation this autumn; last year's consultation round drew just under 150 responses.

Sources: ICO, The Next Web, Regjeringen.no, IMY

Anthropic reported on October 16 that Claude Mythos 5, working in a closed loop in which an automatic evaluator checked and scored every proposal, designed quantum circuits for eight tasks that the company says are more efficient than previously published constructions. The work was led by Andi Gu, with co-authors from Harvard and QuEra Computing.

The results include an elliptic-curve point-addition circuit that uses 12 percent more qubits but 17 times fewer gates than the smallest earlier design, and a Reed–Solomon decoder with roughly twelve times fewer gates than Google's earlier construction. These are designs and resource estimates, not runs on a quantum computer, and the method only works for circuits whose output can be checked efficiently. The figures come from Anthropic itself, and the work is available as a preprint on arXiv.

For the Nordics: The sources say nothing about the Nordics, so this is an assessment. Lower gate costs reduce the demands on future quantum computers, which matters for the Nordic groups building or researching such machines, including Chalmers and WACQT (Wallenberg Centre for Quantum Technology), KTH and NTNU. Vinnova has just proposed quantum technology as one of seven strategic acceleration programmes, and AI-designed quantum circuits add a new question to that discussion: how should results that machines have produced be reviewed?

Sources: Anthropic, Vinnova (via Hurbra)

The Anthropic Institute published a report on October 16 on how employment is changing inside Anthropic itself. The junior share of technical hires fell from 27 to 9 percent, and juniors' share of all technical staff fell from 21 to 14 percent since early 2025.

According to the report, AI agents worked about six times as many hours in August as the employees directing them in interactive sessions, and in April about a third of machine-learning managers and a fifth of software-development managers said they were hiring less because of AI. The authors caution that the tools were not adopted at random, that those who adopted the tools were already on a steeply rising productivity trajectory, and that the material comes from a single AI company whose ways of working cannot be transferred directly to the wider economy.

For the Nordics: The sources say nothing about the Nordics, so this is an assessment. The report concerns one AI company and says nothing about the Swedish or Norwegian labour market, but it echoes a worry that entry-level jobs disappear first. Whether it shows up in Sweden and Norway will have to be settled by register data, for example from Statistics Sweden (SCB) and Statistics Norway (SSB).

Sources: Anthropic, World Today News

Nscale's data centre in Glomfjord, Norway, is about six months behind schedule according to Bloomberg, with leaks in the cooling system and a transformer fire, and 6.4 of a planned 30 megawatts are operational. The figures come from Bloomberg's report, which draws on people close to the project and public records. Data Center Dynamics, which relays them, has asked Nscale for comment and verification.

Meanwhile Nscale is preparing an IPO and, according to the reporting, wants a valuation of up to $35 billion, against $14.6 billion at its latest private round in March. According to Bloomberg, as relayed by Crypto Briefing, several investors are pushing back on that valuation. Contracts worth $103.4 billion, mainly with Anthropic and Microsoft, make up the backlog, but only about 2.5 percent of the contract value was active at the end of August. The loss for the first half of 2026 was $1.02 billion on revenue of $140.6 million.

For the Nordics: The facility is in Glomfjord in Nordland and is a large-scale Norwegian AI project. The sources say nothing about consequences for Norwegian authorities or municipalities, so this is an assessment: delays and uncertain financing can hit local business and Norwegian ambitions to host large AI capacity, and remind Nkom, Statnett and the municipalities that capacity on paper and capacity in operation are different things.

Sources: DCD via BeBeez, Crypto Briefing

OpenAI CEO Sam Altman confirmed on the Sources podcast that the company will build a humanoid robot, and told Fortune that he expects a demo in 2027. According to him, robots walking city streets are a few years further off.

Altman justified the form by saying the world is built for people, and said the robot's control system, its “brain”, matters more than the body. OpenAI has not said what the demo will show, what the robot will look like or when outsiders will get access to one. The company's careers page lists 27 robotics openings, and the postings point to custom-built actuators and to roles in manufacturing and supply chain. In the short term the company wants general-purpose robots that help skilled workers build infrastructure.

For the Nordics: The sources say nothing about the Nordics, so this is an assessment. The Norwegian-founded humanoid company 1X Technologies operates in the same market and Denmark's Universal Robots in an adjacent one, and Sintef and NTNU conduct robotics research. OpenAI's move into hardware strengthens the picture that competition is decided by the AI software as much as by the body, which is a question for Nordic manufacturing, a large buyer of robotics.

Sources: Engadget, AI Weekly

DeepSeek released V4.1-Flash-1017 on October 17, a new post-training version of V4.1-Flash available in the company's app, on the web and through its API, where it is called deepseek-flash. The model is a mixture-of-experts system with a one-million-token context window, and the company is opening search and fetch APIs at $0.75 per thousand successful requests.

By DeepSeek's own figures, 1017 beats the earlier version on all reported benchmarks, by 2.0 to 11.1 points, and scores 77.0 on DeepSWE against 74.0 for Claude Opus 5. Opus 5 still leads on, among others, Terminal-Bench 4.0 (51.8 vs. 42.3) and HLE without tools (56.3 vs. 42.6). The figures come from the company itself and are not independently verified, and the sources differ on whether the weights for the 1017 version specifically are open – the weights for the September version are MIT-licensed.

For the Nordics: The sources say nothing about Nordic users, so this is an assessment of consequences, not a reported fact. Low prices and long context may appeal to Swedish and Norwegian developers, but an API run by a Chinese company raises questions about which personal data leaves the EU/EEA, where IMY and Datatilsynet are the supervisory authorities. If the weights are open, the model can instead be run in-house or at a Nordic or European provider; the license should be checked before building it into a product.

Source: MarkTechPost, AI Weekly

OpenAI published 722 mathematical manuscripts on October 6, reportedly produced by an internal frontier model that was given about 4,000 problems to work on. On October 15 the Simons Foundation announced an initiative to examine and build on AI-produced results, to which OpenAI is contributing $5 million.

The results are at different stages of verification: OpenAI shares formalizations in the Lean proof assistant for many of the proofs but not all, and three manuscripts were reportedly withdrawn after a sign error while 14 were revised. The most talked-about result lies close to the Riemann hypothesis but is not a proof of the hypothesis itself. Several mathematicians have told The Verge that a full review could take years. The Simons initiative is a partnership between the Simons Foundation, Anthropic and OpenAI and totals $15 million in funding and resources, according to the foundation; how much Anthropic contributes has not been disclosed.

For the Nordics: The sources say nothing about the Nordics, so this is an assessment. The question of how machine-produced mathematics should be reviewed also reaches Nordic mathematics departments, including at KTH, Chalmers, the University of Oslo and NTNU. In Norway the Abel Prize is awarded by the Norwegian Academy of Science and Letters, which makes the assessment of AI-produced proofs a question for Norwegian prize institutions too. How the Research Council of Norway and the Swedish Research Council will treat AI-assisted results in grant applications is still open, and we do not know whether the results hold.

Source: OpenAI, iNyheter, Simons Foundation

On October 12 Vinnova presented the report Roadmaps for Strategic Technologies (Färdplaner för strategiska tekniker), in which the agency proposes seven state acceleration programmes: AI, 6G, semiconductors, quantum technology, materials and production, biotechnology and space technology. The AI programme, AI Acceleration Sweden, is proposed to start in January 2027 with at least SEK 1 billion in state funding in its first year.

The other programmes would normally receive at least SEK 200 million a year from 2028 and at least SEK 1 billion in state funds from 2032. Vinnova also proposes a national investment fund with SEK 75 billion in state capital and up to SEK 150 billion in guarantees, and that state R&D investment reach at least 1 percent of GDP in 2030 (0.77 percent in 2024). These are proposals: the money has not yet been decided by the government.

For the Nordics: The proposal concerns Sweden. The sources name no individual institutions, but in our assessment it affects, among others, Chalmers and WACQT (Wallenberg Centre for Quantum Technology), KTH and RISE, as well as the Wallenberg foundations' AI and quantum efforts, including WASP. The decision rests with the government and ultimately the Riksdag. The Research Council of Norway and counterparts in Denmark and Finland get a point of comparison when their own AI and quantum investments are discussed.

Source: News Øresund, Hurbra

Norway's Venstre party wants to introduce a tax deduction scheme, called Kompetansefunn, for companies that invest in their employees' skills instead of replacing them with AI. The proposal was put forward by party leader Guri Melby in connection with Venstre's national conference in Lillestrøm on Saturday, October 17, according to Adressa/NTB.

The idea is to change the calculation so that retraining and new tasks become more profitable than swapping staff for AI. Melby pointed out that many of the jobs disappearing are entry-level positions, and warned that if juniors disappear now, companies will lack senior staff in ten years. Adressa cites DNB and Posten Bring as examples of cuts where AI is given as a contributing factor; on October 6 DNB announced that the bank is cutting 400 full-time positions as its technology and services division is reorganized.

For the Nordics: This is a party proposal and not decided policy, but it touches on an issue under way across the Nordics. Finansforbundet has questioned whether the cuts can be carried out properly before New Year, and the proposal highlights how the gains from AI should be shared between owners, customers and employees, a question that concerns the Storting, NHO and LO. We have not seen a corresponding proposal in Sweden.

Source: Adressa, Fagbladet

Anthropic has confirmed a long-term lease for capacity in EverWind's planned data centre at Point Tupper in Nova Scotia, the company's first data centre in Canada. EverWind says the investment totals C$10 billion, of which C$8 billion is for the data centre and C$2 billion for onshore wind power.

The first 95 MW phase is due to be operating in late 2027, with a 325 MW expansion targeted for 2028, subject to regulatory approval, and 420 MW at full build-out. The project is expected to create around 2,000 construction jobs and 215 permanent positions, with a closed-loop cooling system and several hundred megawatt-hours of battery storage. Nova Scotia NDP leader Claudia Chender has called for regulation, and a consortium linked to Membertou First Nation is negotiating a possible ownership stake.

For the Nordics: The development shows how AI companies are seeking power and land outside the United States, and Nordic places such as Luleå and Mo i Rana, where a data centre is now being converted into an AI facility, compete for the same kind of investment. Questions about the grid, renewable power and local benefit, which Svenska kraftnät and Statnett also handle, decide which projects get built. The sources say nothing about the Nordics, so the connection is an assessment.

Source: Global News, US News/Reuters

Meta Superintelligence Labs released the weights of Muse Spark Open on October 15, a multimodal model with 816 billion parameters, 42 billion of them active per request, and a context window of one million tokens. Meta calls it the most capable US open-weight model to date.

The model ships under Meta's own Muse license, which has limited commercial terms that Meta says let many developers use it for free. Meta says it runs on two Apple Mac Studios with M5 Ultra chips via the MLX framework. The weights can be downloaded from, among others, Hugging Face, Amazon SageMaker, Google Cloud Model Garden and Microsoft Foundry, and APIs are offered by providers including Together AI, Fireworks AI, Groq and OpenRouter. Meta's own comparisons have not been independently reviewed.

For the Nordics: The sources say nothing about Nordic users, so this is an assessment rather than a reported fact. Open weights let Swedish and Norwegian organizations run the model in-house or at a European cloud provider, which can matter for public bodies and regions that do not want to send data to a US API and that fall under supervision by IMY and Datatilsynet respectively. The license has limited commercial terms, so they should be read before the model goes into a product.

Source: Meta, The Neuron

Sferical AI presented the Sferical One supercomputer in Stockholm on October 15, a shared Nvidia-based platform for AI development that its founders AstraZeneca, Ericsson, Saab, SEB and Wallenberg Investments will use together with partner SKF. The system has been running since late summer, so the ceremony was a formal presentation.

According to Nvidia's technical overview, the facility consists of 16 racks with 1,152 GPUs based on DGX GB300, and the companies plan to build services on Nvidia's open Nemotron models. AstraZeneca is aiming at drug discovery, Ericsson at AI models for telecom networks, Saab at defense technology and simulation, SEB at data analysis and software development, and SKF at predictive maintenance and quality control. Nvidia CEO Jensen Huang and Deputy Prime Minister and Minister for Energy, Business and Industry Ebba Busch attended. Sferical AI CEO Jenny Nordlöw calls it a sovereign platform, but Nordisk Post points out that the project still depends on technology from a US company and that the benefits are not yet proven at scale.

For the Nordics: This is a primary Nordic story. AstraZeneca, Ericsson, Saab, SEB and Wallenberg Investments are pooling compute in a shared platform for training and running AI models, while the dependence on Nvidia remains. The sources say nothing about outside customers or other Nordic countries, so it remains to be seen whether the platform will open to more than its founders.

Source: Nordisk Post, TV4/TT

Tillväxtanalys, the Swedish Agency for Growth Policy Analysis, and Uppsala University showed in a register study on September 30 that recent graduates from AI-exposed programs find it harder to get established in the labor market than other recent graduates. Their earnings have developed more weakly, and the share working in the most AI-exposed firms has fallen.

The study draws on register data up to summer 2026. The researchers see no sign of a major transformation of the labor market as a whole, and job mobility is unusually low, especially in the AI-exposed parts of the economy. Productivity has risen in firms that use AI. Co-author Oskar Nordström Skans, professor at Uppsala University, says according to SvD that no very large transformation is visible yet but that things have gone somewhat more slowly for those with a university degree; causation is uncertain. Doctoral student Leo Sokolow Romin calls the changes worrying if they continue.

For the Nordics: This is a primary Swedish story with a Norwegian example. On October 6, DNB announced that it can cut its workforce by about 400 full-time positions; CEO Kjerstin Braathen says AI is changing how the bank works and that it sees significant efficiency gains. The study does not cover DNB or Norway, and the sources do not say how much of DNB's reduction is due to AI.

Source: Tillväxtanalys via TT, CIO Analytics (SvD)

A study in Nature, presented by the US National Institutes of Health on October 14, used AI to analyze PET-CT scans from over 100,000 adult cancer patients and linked detectable brown fat to lower rates of several diseases. The research team is led by Paul Cohen at Rockefeller University.

According to News-Medical, which reports the study's figures, the odds among the 8,709 of 101,243 patients with detectable brown fat were 58 percent lower for type 2 diabetes and 47 percent lower for chronic ischemic heart disease, and detectable brown fat was also linked to lower odds of depressive episodes and anxiety disorders. Patients taking angiotensin receptor blockers had higher brown fat activity. The blood pressure drug losartan raised activity in cultured brown fat cells and energy expenditure in mice while body weight stayed steady, but the effect in humans has not been tested. The study is retrospective and shows associations, not cause and effect.

For the Nordics: The sources name no Nordic party, so this is an assessment. The material comes from cancer patients examined with PET-CT and cannot be transferred directly to Swedish or Norwegian patients, and losartan's effect on brown fat in humans remains to be shown in clinical trials. The Nordic countries' broad health registers make prospective follow-ups conceivable, but that is a possibility and not something the sources report.

Source: NIH, News-Medical