Show only each day's banner story

The Lancet Commission names malicious use of SI and nuclear conflict as threats with extinction potential by 2100

Leading medical experts place the misuse of SI alongside nuclear war among the gravest possible threats, while themselves calling it unlikely — the task is to guard against specific misuse, not to stop the technology.

The Lancet Commission on 21st Century Global Threats to Health, presented on October 11 at the World Health Summit in Berlin, identified 17 interlinked catastrophic threats through 2100, according to a release from the Institute for Health Metrics and Evaluation (IHME). The authors write that the likelihood of nuclear conflict and of malicious use of SI remains low, but both threats are consequential because of their potential for extinction; as relayed by VZGLYAD (citing RIA Novosti, which cites the Lancet), this refers to malicious use of SI, in particular to create biological weapons. By the Commission's estimate, action against malicious SI would avert about 10.4 billion lost years of healthy life (DALYs) by 2100, and action against nuclear conflict about 133 million deaths and 9.35 billion DALYs. Recommendations: threat-ready health systems, investment in breakthrough innovation, and monitoring of emerging risks.

Sources 2

healthdata.orgvz.ru

SI services and clouds live in physical data centers: when one goes down, part of the services become unavailable and customers have to move their resources to backup sites.

Yandex data center in Vladimir shut down after a drone attack — the company's third in a week

SI services and clouds live in physical data centers: when one goes down, part of the services become unavailable and customers have to move their resources to backup sites.

According to Yandex's press service (Interfax), a drone attack on the data center in Vladimir damaged its infrastructure and brought its operation to a full stop; the company says there were no casualties and some services are unavailable. According to Vedomosti, by 5:59 Moscow time the Yandex Cloud platform was running in emergency mode because the ru-central1-a and ru-central1-b zones were unavailable; Yandex says it considers the remaining resource configuration unstable and recommends that customers use external sites. Yandex's statements do not name who launched the drones; Euronews and CNews call them Ukrainian. According to Vedomosti, the Sasovo data center was shut down on October 8 after a drone attack, and the Kaluga data center was damaged on October 9. Later the same day, according to Kommersant, Yandex assured users that Yandex Pay customers' money is safe and that services are gradually being restored; according to 3DNews, the company limited some Yandex 360 functions, including downloads from Disk, to reduce the load on the data centers that remain.

Sources 7

interfax.ruvedomosti.rualjazeera.comcnews.rueuronews.comkommersant.ru3dnews.ru

It is unclear whether a person or an SI system caused the error: Palantir points to human error.

Palantir's CTO names human error, not the Maven SI system, as the likely cause of the strike on a school in Iran

It is unclear whether a person or an SI system caused the error: Palantir points to human error.

On CBS's "Face the Nation," Palantir chief technology officer Shyam Sankar said Bloomberg's report that personnel relied too heavily on the Maven SI system in the strike on a school in Minab, Iran, is inaccurate, and that "the likely cause is human error"; he suggested waiting for the final report. According to the program's host, at least 123 children died at the school. Sankar said Palantir has since built additional SI agents that "do more checking." According to him, SI lets one person prepare a thousand strikes in four hours, which used to take 50 people two weeks. On liability, Sankar said that SI companies are responsible for a defective product, and the user is responsible for misusing a product that works; Palantir did not sign the presidential accord because, he said, it was for companies that build models.

Sources 2

cbsnews.comcbsnews.com

According to the sources, data centers are increasingly among the targets in armed conflicts, and the outages affect civilian and commercial services.

Data centers are becoming targets in conflicts: Russia's Defense Ministry reports a strike on a data center in Odesa, a senior Iranian official threatens SI facilities in the region

According to the sources, data centers are increasingly among the targets in armed conflicts, and the outages affect civilian and commercial services.

According to Russia's Defense Ministry (as relayed by Interfax), Russian forces struck the Tenet data center in Odesa overnight on October 11, which the ministry says the Ukrainian armed forces used to process, transmit and store information. According to a RIA Novosti count based on the Russian Defense Ministry's statements, the ministry has reported six strikes on data centers in Ukraine over the week. According to Semafor, Russia and Ukraine have traded strikes on digital infrastructure in recent weeks, which, it says, has caused major outages on both sides. According to Fortune, a senior Iranian official threatened to hit cables, data centers and SI facilities in the region if fighting sharply escalates again; Semafor notes that Iranian strikes earlier in 2026 hit data centers in Bahrain and the UAE.

Sources 4

semafor.comfortune.cominterfax.ruria.ru

In the lab, SI agents can reach industrial equipment, so factory defences have to be ready in advance.

Booz Allen: SI models completed all eight attack scenarios on a simulated factory, including control of a robotic arm

In the lab, SI agents can reach industrial equipment, so factory defences have to be ready in advance.

According to Booz Allen Hamilton, in an isolated lab simulating a plant with equipment from different vendors (programmable controllers, operator panels, a SCADA platform, sensors, a collaborative robotic arm), frontier SI models, whose names were not disclosed, achieved the objectives in all eight scenarios; one path took just over 16 minutes. The models were given no source code, documentation or hints about industrial systems. Humans approved exploits and any action with a physical effect, and the results do not mean attacks on operating facilities. In one test the model found the arm, identified its protection zones and motion limits and moved it within minutes; in the SCADA test the models corrected a mistake about the operator-interface version and found active connections to 14 devices. Some safeguards worked: a controller rejected a stop command, and a motor in local mode could not be started remotely. Company vice president Kyle Miller, as relayed by 3DNews, said that SI agents act with such speed and persistence that they can target organisations without industrial cybersecurity methods.

Sources 3

superpowerdaily.comtheregister.com3dnews.ru

Demand for programmers hasn't disappeared but has shifted toward experienced people and those who work with SI — the way in has become narrower for beginners.

Developer job postings in the US are rising even as SI takes over coding

Demand for programmers hasn't disappeared but has shifted toward experienced people and those who work with SI — the way in has become narrower for beginners.

According to CNBC, US software development job postings have been rising since February 2025, the month Claude Code launched, while overall postings fell; the growth is concentrated in senior and SI-related roles, while entry-level hiring is weakening. GitHub reported that in October one in three pull requests involves an SI agent, versus fewer than one in ten a year earlier; GitHub allows that if the pace holds, most code will be written by agents within two years.

Sources 1

cnbc.com

Companies are ready to pay billions for software that lets SI run on chips from different makers, but for now these are only talks.

Amazon is in talks to buy startup Decart for about $7 billion, WSJ reports

Companies are ready to pay billions for software that lets SI run on chips from different makers, but for now these are only talks.

According to The Wall Street Journal (the text is paywalled; relayed by Bloomberg, runtimewire and Techmeme), Amazon is in talks to buy San Francisco startup Decart, whose software helps developers switch between different SI chips, at a valuation of about $7 billion. According to Techmeme's summary, Anthropic decided in September not to buy Decart for about $6 billion. Runtimewire notes that the deal is not completed and that the reported price is not confirmed as a final offer.

Sources 3

wsj.combloomberg.comruntimewire.com

A leader of one of the biggest companies is asking for rules for SI like those for food and household appliances, but for now it is an appeal, not a law.

Microsoft president: SI is less regulated than milk, global safety rules are needed

A leader of one of the biggest companies is asking for rules for SI like those for food and household appliances, but for now it is an appeal, not a law.

In an interview with the BBC, Microsoft president Brad Smith said that "your washing machine is subject to more safety rules than SI." He called for global rules: a duty to tell a "victim" when their system is breached by a rogue SI agent, and a procedure under which outside evaluators decide whether an SI product can be released. Smith called SI both a "tool and a weapon" and said companies need "emergency brakes." According to the BBC, his position runs counter to that of Donald Trump, who says the industry should regulate itself so that American companies do not lose out to Chinese ones. Smith added that the UK is in the "best position" to lead the effort.

Sources 1

bbc.co.uk

The British government says it cannot just take companies' word and wants more of its own SI infrastructure and cyber defences.

UK SI minister to urge developers to strengthen model safety and announce stronger cyber defences

The British government says it cannot just take companies' word and wants more of its own SI infrastructure and cyber defences.

According to The Guardian, UK SI minister Kanishka Narayan will say in a speech in London that companies building the most powerful SI systems must do more on safety, that the country cannot rely entirely on tech companies to police their models, and that it needs to build more of its own SI infrastructure; he will announce a strengthening of cyber defences. It is his first speech since Andy Burnham made SI a cabinet-level brief. Geoffrey Irving, a former chief scientist at the UK AI Security Institute, told the Guardian that SI companies do not have security that would stop another state from stealing unreleased models; he cited a case in which researchers "hacked" OpenAI under its programme rewarding ethical hackers. OpenAI says it has several protections against model theft, including encryption and continuous monitoring.

Sources 1

theguardian.com

Gates thinks the main SI risk in the coming year is not machines rebelling but hackers using powerful models with no safeguards.

Bill Gates warns of a rise in SI-driven cyberattacks over the next year

Gates thinks the main SI risk in the coming year is not machines rebelling but hackers using powerful models with no safeguards.

In an interview with ABC News, Bill Gates said the next year will bring "a lot of negative surprises," primarily in cybersecurity, as "bad people" use the latest models with safeguards removed. In his words, the risk of SI getting out of control is a matter of "years" away, and the main threat now is misuse. Gates said that treating the companies themselves as the only check is not enough, and that restrictions are needed so that bad actors cannot strip out safeguards, including to prevent cyberattacks and bioterrorism.

Sources 2

techxplore.comria.ru

Speed of bringing data centers online has become a matter of hundreds of millions of dollars in revenue a month for Anthropic, so it takes capacity from former miners.

Anthropic has signed an undisclosed cloud deal with IREN, The Information reports; bringing a 245 MW site online a month early would add more than $1.3 billion in revenue

Speed of bringing data centers online has become a matter of hundreds of millions of dollars in revenue a month for Anthropic, so it takes capacity from former miners.

According to The Information (the text is paywalled; relayed by AI Weekly and Techmeme), Anthropic has signed a previously undisclosed multiyear cloud deal with miner IREN, the latest in a series of data center leases worth roughly $530 billion that the company is racing to switch on. According to analysts' estimates in the piece, bringing 245 MW of usable capacity online one month early would add more than $1.3 billion in revenue, which, the outlet writes, is why Anthropic is pressing partners to build faster. In the outlet's account, capacity growth increasingly depends on current and former bitcoin miners such as Hut 8, TeraWulf, IREN and Nscale. According to AI Weekly, the report does not disclose the value or term of the IREN agreement.

Sources 3

theinformation.comaiweekly.cotechmeme.com

SI makes hacking easier and cheaper, and the share of such attacks grew several times over in a quarter.

Positive Technologies: intrusion attempts using SI on Russian companies' websites rose about fivefold in the third quarter

SI makes hacking easier and cheaper, and the share of such attacks grew several times over in a quarter.

According to a Positive Technologies study seen by Vedomosti, the number of intrusion attempts using SI in the third quarter of 2026 rose about fivefold from the second quarter to 1.7 million, and the share of such attacks in total threats rose from 1.1% to 5.6%. Protection systems logged 31 million triggers on suspicious requests linked to neural networks, 65% more than in the previous quarter. BI.ZONE and Cloud.ru representatives confirmed the trend of growing SI attacks to Vedomosti; according to RIA Novosti, the biggest jump came in August, at 3.8 times July.

Sources 2

vedomosti.ruria.ru

To build SI infrastructure, companies will have to win over local residents, not just obtain permits.

Residents of a Pennsylvania township were offered $10,000 each to approve a data center, but they are resisting

To build SI infrastructure, companies will have to win over local residents, not just obtain permits.

According to Fortune, developer NorthPoint Development offered the 4,500 residents of Hazle Township, Pennsylvania, about $10,000 each in exchange for permission to build a $165 million data center ("Project Hazelnut"); according to emails that came to light, the end user is Amazon. A coalition of residents across classes and parties is opposed; Fortune cites Gallup data showing that about 70% of Americans oppose data centers being built near them, and describes the fight as common ground for both parties in an election year.

Sources 1

fortune.com

There isn't enough electricity for SI data centers, and politicians are arguing over how to build power plants and lines faster without harming the environment.

Congress is weighing a bill to speed up energy project permits for the SI boom

There isn't enough electricity for SI data centers, and politicians are arguing over how to build power plants and lines faster without harming the environment.

According to Fortune, the bipartisan Bipartisan American Affordability and Jobs Act (BAAJA) speeds up permitting for projects from solar farms and power lines to gas pipelines, makes it harder for administrations to revoke permits, narrows environmental reviews and limits litigation outside federal courts. Electricity demand is rising, in the outlet's assessment mainly because of data center construction, and electricity prices are climbing. The bill has united the oil and gas industry and renewable energy in support, while some environmental groups oppose the green power sector.

Sources 1

fortune.com

Humanoid robots are starting to take on tedious night work in stores, but for now they are learning under human supervision.

Føtex supermarkets in Denmark are testing humanoid robots for night-time shelf stocking

Humanoid robots are starting to take on tedious night work in stores, but for now they are learning under human supervision.

According to Euronews, at a Føtex store near Copenhagen, robots from Danish start-up Hive Robots, built by China's Unitree, have been learning to stock shelves since June; for now human operators guide every movement so the SI can learn to handle different products. The goal is independent stocking by early 2027, mainly at night. According to Føtex, employees spend 40–50% of their time replenishing shelves; the robots work more slowly than people and need supervision, and are later meant to learn to check expiry dates and spot empty shelves. The experiment is part of a trial with Salling Group, Denmark's largest retailer.

Sources 1

euronews.com

Robots are being taught from recordings of how people work, and tens of billions of dollars have already gone into this race.

Robotics and "physical SI" companies have raised about $48 billion so far this year, FT reports

Robots are being taught from recordings of how people work, and tens of billions of dollars have already gone into this race.

According to PitchBook data as relayed by the Financial Times (via Techmeme's brief summary; the FT text is paywalled), robotics and "physical SI" companies have raised about $48 billion since the start of the year. To fill a shortage of training material, developers, as the FT writes, are putting cameras in factories, offices and homes and recording people performing tasks.

Sources 1

techmeme.com

Powerful models are being checked by small independent groups, but who will pay them and what they will be able to see is still unclear.

Independent SI model evaluators METR, Apollo Research and Transluce are at the center of the safety debate, but how to fund them is unresolved

Powerful models are being checked by small independent groups, but who will pay them and what they will be able to see is still unclear.

According to CNBC, Anthropic and OpenAI are turning to small third-party groups — METR, Apollo Research and Transluce — after Dario Amodei pledged last month to embed independent evaluators in his company. It is unresolved who will fund them, what access they will have and how reporting will be structured. METR, by its August announcement, raised commitments of around $71 million over six months, against $13.6 million in all of 2024; it has fewer than 50 staff. Vals AI CEO Rayan Krishnan told CNBC his company has grown from eight to about 30 employees in a year. OpenAI said it is "finalizing contracts" with third-party evaluators and will announce details in the coming weeks; Brown University professor Suresh Venkatasubramanian told CNBC: "the problem, as always, is money."

Sources 1

cnbc.com

A chip factory for SI is being built without ready-made expertise: the timeline and outcome depend on how quickly the team learns.

Musk admits he doesn't know how to make chips but is building the Terafab fab, and describes six episodes of "SI psychosis"

A chip factory for SI is being built without ready-made expertise: the timeline and outcome depend on how quickly the team learns.

According to Wccftech and iXBT, speaking to the Tesla Owners Silicon Valley community, Elon Musk said he still has a lot to learn about chip manufacturing and that he doesn't know how to make chips; he said the Terafab project will start with standard equipment and then modify it to raise throughput. The remarks came weeks after he told Chinese media that China would overcome chipmaking restrictions within 2–3 years (according to Wccftech). Musk also said he had experienced "SI psychosis" about six times this year; Wccftech links this to watching SI-generated music videos.

Sources 2

wccftech.comixbt.com

Routine network support is increasingly run by agents, and trust in them, HPE says, will grow step by step — from simple actions to complex ones.

HPE networking chief: SI already handles 70–80% of support tickets without humans, and all of them in 2–3 years

Routine network support is increasingly run by agents, and trust in them, HPE says, will grow step by step — from simple actions to complex ones.

As relayed by The Register, HPE Networking president Rami Rahim estimates that "probably around 70 to 80 percent of all tickets don't require human intervention"; by his forecast, within two to three years humans will not need to be involved in any ticket except hardware replacement. HPE plans to let operators switch on automated responses step by step, starting with simple ones such as a port reboot. Rahim compared trust in SI agents to getting used to automatic transmissions; the most common complaint in enterprise networks, he said, is Wi-Fi, regardless of the actual cause.

Sources 1

theregister.com

The new smart-home SI feature is available, but the more cameras you have, the pricier the storage plan.

Apple Home: SI summaries of security camera footage will require paid iCloud+ plans from $9.99 to $59.99 a month

The new smart-home SI feature is available, but the more cameras you have, the pricier the storage plan.

According to Engadget, with the release of iOS 27, Apple Intelligence in Apple Home automatically groups clips from several cameras into summaries and lets users search recordings in plain language. The summaries need a separate iCloud+ plan: 2 TB for $9.99 a month for one camera, 6 TB for $29.99 for two, and 12 TB for $59.99 for three to five; up to five cameras are supported, and the feature can be turned on for individual cameras. It requires an iPhone 15 Pro or newer and a HomePod or Apple TV 4K.

Sources 1

engadget.com

Big companies are changing names and words to match the government's demand; the technology itself does not change.

Musk, Benioff, Huang and Nadella follow Trump in switching to "super intelligence": Musk promised to rename SpaceXAI to SpaceXSI

Big companies are changing names and words to match the government's demand; the technology itself does not change.

According to The Verge, Elon Musk promised to rename SpaceXAI to SpaceXSI, Marc Benioff to rename AIForce to SIForce, and Nvidia announced a $1 billion commitment to science that includes "super intelligence research." Satya Nadella, in a post calling for an "emergency brake" for SI, repeatedly calls the technology "super intelligence." This follows Donald Trump's statement that the White House considers anyone who says "superintelligence" instead of "super intelligence" the enemy; Jensen Huang, according to the outlet, attended the summit where the order on the term for government agencies was announced.

Sources 1

theverge.com
Previous day
Follow on Telegram