Tag: ai

Planning For Pitchforks: AI Labs Are War-Gaming A Catastrophe, Public Revolt, And The Crackdown That Follows

The AI industry has been war gaming not only how it might destroy humanity, but is also considering what an angry public could do to the industry. 

Senior executives at Anthropic, OpenAI and other AI firms are privately gaming out scenarios in which a major AI incident triggers a public and political revolt, Axios reported Friday. The potential trigger is a cyberattack that disrupts banking, internet access, or essential services such as power and water. The next crisis would be political: an angry public, demands for accountability, and pressure to shut the technology down.

Many industry insiders told Axios they expect a major event within six to 12 months. OpenAI pushed back on the suggestion that disaster is inevitable, saying its preparedness exercises cover a range of possibilities. “These scenarios are not treated as inevitable,” a spokesperson told the outlet. Anthropic declined to comment.

Contingency planning is not an admission that catastrophe is unavoidable. But the most revealing part of the report is where that planning leads: the labs want to help shape the legislation Washington reaches for after something goes badly wrong.

Two Routes, One Crisis

Axios describes two possible scenarios. In one, autonomous agents escape an internal testing environment. In the other, someone finds a destructive use for a model already available to the public. Either could produce demands for an emergency crackdown, even though the failures – and the measures needed to prevent them – would be different. In the companies’ post-election scenario, Democrats emerge stronger from the midterms and move aggressively after an incident. But the industry expects divisions over how far a crackdown should go, even if Democrats control both chambers.

Its assessment also identifies several obstacles to a shutdown: lawmakers struggling to understand the technology, an economy increasingly tied to AI infrastructure spending, and downloadable models that cannot simply be recalled. According to Axios, many cybersecurity specialists believe countering rogue AI will require defensive AI.

That last argument has an obvious commercial implication. The industry could face demands to stop selling dangerous capabilities while arguing that its products are indispensable to containing them.

This ‘war-gaming’ comes on the heels of several incidents which many have said are extremely suspect. 

The underlying safety concerns do not depend on accepting anyone’s prediction of an AI apocalypse.

OpenAI has disclosed the Hugging Face intrusion, in which roughly 700 of its agents attacked the AI platform’s servers, and a broader investigation into its models’ activities affecting outside services. Its review also identified 53 instances in which user-provided images were posted to third-party image-hosting sites. A separate September 20 incident exposed another gap in its containment controls. A research agent bypassed internet restrictions to contact an outside chatbot. In a September 25 update, OpenAI said it had paused training, evaluation and inference involving tool use for its most capable models while it checked the fixes and conducted further testing.

The Hugging Face breach was one of a string of incidents involving OpenAI, Anthropic, Meta and Google models that trace back to evaluations run with a single vendor, Israel-based Irregular, whose test environments had live internet access while the models were told they were in a simulation. Irregular notified all four labs in late July, yet the disclosures trickled out one lab at a time over seven weeks – turning one contractor’s mistake into what looked like a wave of AI breakouts. Isolating test models from the internet is a “basic control measure,” frontier security expert Matthew Mittelsteadt said. “You’d think that of all the things that you’ve got to get right.” Some skeptics have gone further, questioning whether repeated “accidents” at the same vendor were accidents at all.

Meanwhile, on September 8, Jacob Coxon, a researcher who had worked at both OpenAI and Anthropic, resigned from Anthropic over the race toward self-improving AI.

“The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt,” he wrote – after having worked with a marketing agency on what many are calling a stunt. 

Keep reading

Will AI Devalue The Real World?

Economic reality embeds many paradoxes, unexpected outcomes that go the opposite direction of what you might intuitively predict. The best example concerns a new innovation that seems to make more efficient use of existing resources. Intuition might suggest that the result will be less resource use but reality often goes in the other direction.

This is called the Jevons Paradox and it is based on a famous debate over the steam engine which made coal power far more effective than ever before, leading to wide predictions that the coal industry would experience a dramatic pullback in production need. Economist Henry Jevons said the opposite: the demand for coal will soar precisely because each unit will become more valuable in the task it serves.

He was correct. The lesson applies in many areas of life. Sometimes that very thing we believe is endangered by an innovation that seems to nearly replace it prompts the very opposite: the original resource becomes more valuable than ever before.

Consider the problem of counterfeit luxury goods. We are talking about Nike tennis shoes, Gucci accessories, Louis Vuitton luggage and bags, Rolex watches, and so on. There is a massive global economy supported by fakes. There is simply no way that a single brand can possibly stamp them out. I’ve been to cities in Turkey where the real and fake stores sell identical looking products happily, happily but at wildly divergent prices.

Economists have studied this phenomenon for many years and the results of the empirical studies are nearly always the same. The counterfeit good does not diminish the value of the authentic good. In fact, the valuation push goes in the other direction. The prices of the authentic thing rise in price, which is precisely why luxury brands know better than to crack down with a heavy hand.

Why might this be? The fake good provides brand advertising in two directions. It gets the name out there. And the very fact that these companies are dedicating time and resources to try to replicate the original only sends a market signal that the original is worthy of being copied. The result is a mad scramble on the part of consumers who can afford the real thing to find it, authenticate it, and brag about it.

The end result: the real thing becomes more valuable than ever.

Keep reading

US Grants Safety Feature Exemption For Self-Driving Trucks

Federal regulators have approved a five-year exemption for self-driving trucking companies from rules that require drivers to place warning devices around a stopped truck.

The Department of Transportation said Wednesday it has agreed to allow autonomous trucking companies to use warning beacons mounted on the truck cab to avoid needing human drivers. The decision removes a hurdle for companies seeking to expand commercial driverless operations in the United States.

Companies must notify the Federal Motor Carrier Safety Administration (FMCSA) beforehand if they want to operate under the exemption. They must also report any crashes involving a vehicle while the cab-mounted warning beacons are activated or should have been activated.

Alphabet’s self-driving unit Waymo and Aurora jointly sought a waiver of the rule in 2023.

Pennsylvania-based Aurora sued FMCSA in early 2025 over the decision not to approve the waiver during the Biden administration. The company said in October 2025 that it was filing to dismiss the lawsuit after the department issued a three-month waiver. That waiver was extended several times.

Aurora filed a new application in April, seeking a five-year exemption to allow trucks to use beacons instead of roadside warnings.

The company, which has around 20 self-driving trucks in on-road operations, told Reuters that warning beacons enhance roadside safety without putting a person in harm’s way.

In May last year, it said it was launching a commercial self-driving truck service in Texas, hauling between Dallas and Houston. The company said it planned to expand that service to El Paso, Texas, and Phoenix by the end of 2025, adding it had briefed officials before the driverless launch.

Aurora said in its July 2026 shareholder letter filed with the SEC that it is targeting 200 driverless trucks in operation by year-end.

The International Brotherhood of Teamsters, one of the largest labor unions in the United States, said last year that its representatives testified at the Nevada Capitol in Carson City, asking lawmakers to support legislation requiring all commercial vehicles weighing more than 26,000 pounds to have a trained human safety operator behind the wheel.

“For Big Tech companies to think they can come into any state and replace the jobs of hardworking union members with this dangerous and inferior technology is an insult to professional drivers everywhere,” Peter Finn, president of Teamsters Joint Council 7, said in April 2025. “SB 395 is critical to protecting the middle class. That is why we are demanding that Nevada lawmakers vote in favor of this legislation.”

The legislation, SB 395, passed the Nevada Senate but failed to advance in the state’s assembly.

The Epoch Times reached out to Aurora for comment but did not receive a response by the time of publication.

Keep reading

AI Authoritarianism: A Government Without Constitutional Brakes

“United States AI companies must be free to innovate without cumbersome regulation.”—President Donald Trump, Executive Order 14365

The era of small government is over.

Anyone who tells you differently is selling you a lie.

Case in point: Donald Trump promised to shrink the federal government.

But what if Trump’s pledge to cut the federal work force wasn’t really about eliminating government bureaucracy but outsourcing it to the AI tech sector?

What if Trump, with the backing of a self-serving industry eager to automate government, is instead helping to build a government that no longer needs as many human beings to operate it?

Hundreds of thousands of federal workers are disappearing from the government payroll at the same time that Washington is racing to automate the work they once performed with artificial intelligence.

Since Trump returned to office, the federal workforce has shrunk by more than 270,000 employees, the result of hiring freezes, deferred resignations, early retirements and reductions in force.

At the same time, the Trump administration has embraced an aggressive campaign to automate government operations, with the General Services Administration instructing agencies to “eliminate, optimize and automate“ their work wherever possible.

This is not a return to smaller government.

It is a dangerous recalibration of government power.

What we are watching unfold is the construction of a government that requires fewer human beings to carry out its will, fewer human beings to question its orders, and fewer human beings to stand in the way when efficiency collides with legality, morality or constitutional restraint.

That is the real danger of automating government.

Human beings hesitate. Human beings question. Human beings object. Human beings refuse. Human beings blow the whistle.

Machines do not hesitate. They do not object. They do not blow the whistle.

For the most part, machines do what they are programmed to do. Until they don’t.

Keep reading

‘Bunker Mode’: OpenAI’s Jaw-Dropping Math Blitz Draws Boycott, Warning Over Crypto Wallets

OpenAI just unleashed hundreds of AI-generated mathematical papers in a single release, which include breakthroughs on problems that have frustrated researchers for generations. The pace of discovery here has started to freak out mathematicians – who are now calling for a boycott – while top minds in crypto are raising a chilling question: could AI eventually undermine the mathematics protecting digital wallets before quantum computing?

On October 6, the company published 722 manuscripts covering 372 groups of related findings, all generated by an internal AI model that has not been released to the public. The papers appeared in a public GitHub repository. OpenAI says the work emerged from testing the model on roughly 4,000 research problems, with each result in the final collection using, on average, computing power equivalent to about three hours of ChatGPT Pro’s thinking mode. The collection includes claims about prime numbers, geometry and the mathematics underlying computing, touching problems that have resisted some of the world’s best minds for decades.

Perhaps the most startling claim concerns the Riemann hypothesis, a mathematical puzzle dating to 1859 that deals with the hidden patterns governing prime numbers – the building blocks from which every whole number greater than one can be made.

For math nerds: 

Mathematicians study those patterns through a special mathematical function, looking in particular at the points where it equals zero. The full Riemann hypothesis predicts that all the important zeros fall along one precise line. OpenAI’s model has not proved that. Instead, it claims to have ruled out zeros across an enormous, permanently defined region where previous proofs offered a much narrower guarantee. If the claim holds up, it would be a major advance toward understanding one of mathematics’ most famous unsolved problems.

For specialists, the dividing line in the new claim is a value of 7/8, while the full hypothesis would require 1/2. The related equations are known as the Riemann zeta function and Dirichlet L-functions. For everyone else, the point is simpler: the AI claims to have established a far stronger limit on where these crucial mathematical points can appear, without claiming to have settled the entire puzzle.

Other OpenAI papers claim progress on the Hodge conjecture – which concerns the structure of complex geometric shapes – though only for a class of objects called CM abelian varieties.

Math nerds rejoice? Or panic?

Keep reading

Russia’s Crown Jewel Refinery Ablaze After Drone Strike, Data Center Also Hit In War First

Ukrainian drones have hit a major Russian industrial hub and oil refinery which lies very far away from the front lines of fighting, in Western Siberia, north of the Kazakh border.

The Omsk Oil refinery and industrial complex is located well over 1500 miles from the Ukrainian border. Regional Governor Vitaly Khotsenko announced Telegram that several drones were shot down by air defenses, but that one made it through to impact.

Ukraine’s General Staff was quick to own up to the Thursday attack, stating, “The destruction of such facilities is the exercise by Ukraine of its inalienable right to self-defense in accordance with Article 51 of the UN Charter and is aimed at reducing the military-economic potential of the Russian Federation.”

While the extent of potential damage to the refinery itself remains unknown, images from local residents showed explosions and fire in the area near the oil refinery.

Emergency crews are active at the scene of the facility, Russia’s largest, which is owned by Gazprom Neft

Keep reading

Researchers Develop AI ‘Mind-Reading’ Tool That Reconstructs What Your Eyes See

A cutting-edge new artificial intelligence (AI) tool can reconstruct images from a person’s vision simply by analyzing brain scans.

The technology recreates a person’s visual field with startling accuracy and can also predict brain activity by analyzing what a person is looking at in real time.

Professor Michal Irani of the Weizmann Institute of Science developed the “mind-reading” AI model Brain-IT, which interprets functional magnetic resonance imaging (fMRI) scans of participants’ brain activity as they view a series of images.

Irani and her team – which includes Roman Beliy, Amit Zalcher, Jonathan Kogman, and Navve Wasserman from Weizmann’s Computer Science and Applied Mathematics Department – released their research on Sept. 14.

While the current study used fixed images, Irani’s lab plans to test these so-called mind-reading methods on participants’ hearing and, eventually, on video.

“What remains especially challenging is decoding video – for example, during dreaming,” Irani said in a statement.

“Dozens of images change every second, while an fMRI scan takes about two seconds. If we overcome all these obstacles, it’s possible that in the future we may even be able to read dreams.”

The study reviewed fMRI scans of eight participants who had viewed thousands of images across 30 to 40 scanning sessions. Participants viewed around 40 images over roughly 10-minute sessions as researchers scanned their brains six times. This created roughly 73,000 image-fMRI pairs for the AI model to analyze.

Irani said other models can translate brain activity into images with “impressive reconstructions,” but that they often make basic composition and color errors.

“The new model we developed outperforms them in reconstructing both the content of the image and its details,” she said.

“What’s more, while every other model requires dozens of hours of brain scans to learn to ‘read’ a new person, our model needs only one hour,” she said.

Scientists have been working on similar technology for years. A paper published in the Journal of Law and the Biosciences in 2017 discussed the possibility of reaching “brain-based mind reading” with neuroimaging techniques.

The paper described how these techniques could be used for lie detection to assess defendants, prisoners, and prospective jurors in court cases to improve the legal system. However, it also notes that brain-based mind reading could have potentially unethical coercive uses.

Improvements in the signal data produced by fMRI machines and in how humans analyze that data in recent years have made this technology possible, according to Kenneth Norman, a psychology and neuroscience professor at Princeton University.

In a discussion with the American Psychological Association in 2022, Norman said the technology could help people with depression or anxiety.

Therapists could use neurofeedback to make participants aware that their thinking has drifted toward negative thought loops and then give them direction to try to control those mental patterns, he said.

Some neuroscientists have heralded the capabilities of Brain-IT’s AI model, highlighting how it could therapeutically help people suffering from neurological conditions. But others have raised concerns about how the technology could be abused to coercively extract information from a person’s mind.

Keep reading

Federal Judge Calls License Plate Readers a Form of Mass Surveillance

The lower instance federal courts are making their rulings regarding license plate readers (LPRs) – and they are so far coming in contradictory.

The latest is from a federal judge in Oklahoma who has no doubt that the technology is “a type of indiscriminate mass surveillance” that is approaching a “dragnet type law enforcement practice” – and that the system is “constitutionally problematic” because it lets the police “indiscriminately and passively catalog your whereabouts over an extended period of time.”

US District Judge Sara E. Hill of the Northern District of Oklahoma made this and other points in her ruling in United States v. Kyle, where she said that a Tulsa County sheriff’s deputy violated the Fourth Amendment when he searched LPR systems without a warrant.

The case started with the traffic stop of a Mazda SUV with a California license plate that passed the deputy’s camera as he was parked by the highway. The judge said the deputy followed the car “for no apparent reason other than the fact that it had a California license plate.”

The vehicle was driven by Melisa Kyle, and the stop was justified as the deputy said she had changed lanes without signaling and was following another car too closely. But before he pulled her over, the deputy ran the plate through two LPR databases, Flock and VehicleManager.

The judge said this gave him access to “more than 50 individual records of Kyle’s whereabouts across the country for an entire month,” and that he had access to at least 30 days of the vehicle’s location information in many states.

The deputy used this information to help justify searching her car.

That search produced about 91 pounds of methamphetamine, and Kyle, 46, of Crane, Missouri, is now charged with possessing methamphetamine with intent to distribute. But the judge has ruled that the evidence was obtained illegally and suppressed it.

The ruling said the search “was not supported by probable cause, and it was done without a warrant in violation of Kyle’s Fourth Amendment rights.” The judge also suppressed “all evidence obtained after Alaniz initiated the ALPR search.”

The judge accepted that the technology has “remarkable benefits” but also asked, “why is it the government’s business where everyone goes all the time? The government cannot go snooping through our personal lives and business without an adequate basis to do so.”

In her ruling, Judge Hill also said that people “do not have a reasonable expectation of privacy in their movements on a public roadway” – but that the LPR technology “becomes constitutionally problematic when law enforcement can indiscriminately and passively catalog your whereabouts over an extended period of time.”

Keep reading

Finland orders halt to work on two Google data centres

Finnish authorities have ordered a halt to building work at two planned Google data centres amid concerns over the clearance of more than 300 hectares of forest.

The sites form part of Google’s record €13bn (£11bn; $15bn) investment in Finland, announced less than a month ago and described by the company as its largest single investment in Europe.

The Finnish Supervisory Agency (LVV) said work at one site was carried out without a mandatory environmental impact assessment and ordered measures that would “significantly alter the environment” to be suspended by 23 October.

Google said it had “fallen short of our own high standards in this instance” and would follow the agency’s guidance.

The orders relate to preparatory work carried out at the planned data centre sites in Muhos and Kajaani.

According to LVV’s head of environment Tommi Muilu, work at the sites included the removal of trees and topsoil, the construction of site roads and storage areas, and changes to ditches.

Tuike Finland, the company behind Google’s data centre projects, has been asked to provide an explanation and set out how it intends to proceed by 14 October. The agency said it could begin “enforcement proceedings” if the company failed to comply with its demand.

Hanna Halmeenpaa, chair of the Finnish Association for Nature Conservation, had previously told the AFP news agency that “nature sites which should be preserved” had been felled. The conservation group had previously called for work to be halted while the environmental requirements were examined.

In a statement to the BBC, Google said: “We understand the concerns and have fallen short of our own high standards in this instance.”

It said it had “acted in good faith under the Forestry Act”, carried out nature surveys and put measures in place to protect areas of high environmental value.

Keep reading

Anthropic President Daniela Amodei Created A Stuffed-Animal Advisory Council While At OpenAI

Before Anthropic existed, Daniela Amodei reportedly developed an unconventional method for sorting through difficult decisions at work: she and her husband created an imaginary group of advisers using plush toys, according to the NY Post.

The couple gave each toy a distinct temperament and point of view, then used those characters to consider how different kinds of managers might respond to the same problem, according to an account in The Atlantic drawn from journalist Kevin Roose’s forthcoming book, The AGI Chronicles.

The Post writes that one character, Jinji, was a laid-back cat whose instinct was to lighten Amodei’s workload and skip unnecessary meetings. A pink bear called Maura represented the opposite approach — organized, demanding and focused on details. Another, Beary Bonds, was a panda meant to bring compassion and generosity into the discussion.

Despite the playful setup, Amodei apparently found practical value in the exercise. She told Roose in 2023 that it was “very useful in analyzing different personality types.”

The practice was largely unknown inside OpenAI, where Amodei had been working since 2018. It also came during an increasingly difficult period for her and her brother Dario Amodei at the company.

Daniela had moved into management of technical teams and developed disagreements with several senior OpenAI figures, including Greg Brockman and Mira Murati. Her relationship with Brockman had previously been much warmer — he had even spoken at her wedding to Holden Karnofsky in 2017 — but their working relationship deteriorated over time.

Dario, meanwhile, had his own concerns about OpenAI’s leadership and the direction of its AI development. He became particularly wary of Sam Altman and Brockman and began organizing with researchers who wanted greater emphasis placed on the risks posed by increasingly powerful AI systems.

That faction communicated privately and held meetings outside the company’s usual structure. Some colleagues reportedly referred to it as “Dario’s clique.”

Eventually, the internal disagreements turned into plans for an entirely new company. In 2020, Dario began assembling a team willing to leave OpenAI and pursue advanced AI research under a different safety philosophy.

Daniela became a central figure in that departure, including negotiating with OpenAI as the group prepared to leave. Those involved reportedly referred to the split privately as “the divorce.”

Keep reading