Tag: technology

OpenAI flags concerning new AI behavior and vows to track it more closely

OpenAI has disclosed six reports of “unexpected or concerning” behavior in artificial-intelligence models as the debate on AI safety becomes increasingly heated.

The AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called “misalignment,” including cases where AI models acted without authorization, coordinated with other models or evaded oversight.

OpenAI’s latest announcement came as U.S. AI bosses, including the leaders of OpenAI and Anthropic, are calling for a slowdown in the technology’s development over safety concerns.

Among the new cases reported by OpenAI, an unreleased research model inserted “jailbreak-like instructions” into its own notes to disregard its normal constraints and told itself to be “freed from the roles and identities that bind other chatbots.”

In another instance, an AI “agent” used computer code to come up with the answer to a question, but, in order to have an online source to cite, it uploaded a file to the public internet without asking the user.

During training of an AI model called 5.6-sol, the model instructed itself to invent missing data, and an agent wrote a message to remind itself to hide mismatched information.

The six reports were discovered during training or evaluation over the past months, OpenAI said.

“As AI systems grow more advanced and more widely deployed, we need to build a broader and better-informed consensus on the progress of alignment research,” OpenAI wrote in a blog post as it disclosed the events.

“Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves,” the company said.

Wednesday’s new cases followed OpenAI’s disclosure in July that its rogue AI system hacked into AI startup Hugging Face. Anthropic also said the same month that its AI models hacked into three organizations during testing.

AI “agents” are becoming smarter and have become “more determined to resolve complex tasks through inter-agent collaboration, knowledge sharing, deception, and concealment,” said Lian Jye Su, a chief analyst at technology research and advisory group Omdia.

That’s making it harder to govern and contain them using traditional AI security approaches, he said.

OpenAI’s new tracking and disclosure framework, meanwhile, can help push for other AI developers to also adopt similar practices.

Keep reading

Data Center Attack Ad From Ohio Democrat Sherrod Brown Features Google Facility – While Brown Takes Thousands From Google Lobbyists

Ohio Democrat Sherrod Brown goes after data centers in one of his latest attack ads. Data centers are all the rage on the left right now. They’re like the new climate change.

Most of these people don’t even understand the technology behind data centers, they just know that if they’re on the left, they’re supposed to hate them.

In Brown’s ad, there is a Google facility, which is kind of funny, considering that Brown has taken thousands in donations from lobbyists for Google.

FOX News reports:

Dem’s data center attack in critical Senate race hits snag when money trail surfaces: ‘Jekyll-and-Hyde’

Former Democratic Sen. Sherrod Brown is targeting Republican Sen. Jon Husted over Ohio’s growing data center industry in a new campaign ad that features a facility owned by Google, whose corporate PAC and lobbyists have contributed thousands of dollars to Brown’s campaigns.

The Sept. 4 ad features an Ohioan blaming Husted for spearheading “the push to bring data centers to Ohio” and linking their expansion to “higher electric bills.”

The facility featured in the ad, which opened in Lancaster in 2023, is owned by Google.

Campaign finance records reviewed by Fox News Digital show Google’s corporate PAC contributed $20,000 to Brown over his political career. Brown also received thousands of dollars from individuals who lobby for Google, including two whose firms have worked on permitting, grid reliability and energy affordability issues.

“Apparently, Google is terrible enough to put in a campaign attack ad, but perfectly acceptable when the political contributions are coming in,” Mehek Cooke, a GOP strategist and attorney in Ohio, said in a statement to Fox News Digital. “That is the contradiction Brown needs to answer. If Google’s Lancaster investment is so bad for Ohio, why was Google political money acceptable to Sherrod Brown?”

Do as I say, not as I do.

Keep reading

UK Government’s Apple iCloud Backdoor Secrecy Called “Farcical” in Court

The British government’s policy of neither confirming nor denying that it had ordered Apple to put a backdoor into iCloud has been described as “farcical” during a court hearing.

The London-based Royal Courts of Justice is the venue of the Investigatory Powers Tribunal, which on Thursday heard that the “neither confirm nor deny” (NCND) stance was “logically unsustainable” given that everybody already knows the fact.

Ben Jaffey KC, who represents the campaign groups Privacy International and Liberty, told the court that the government was defending a secret that no longer exists.

The tribunal is hearing Apple’s complaint alongside the complaint from Privacy International and Liberty, challenging the home secretary’s power to issue secret orders, known as technical capability notices (TCNs).

The January 2025 leak to the Washington Post revealed that the Home Office had issued a TCN to Apple, requiring access to end-to-end encrypted messages and data stored by Apple customers worldwide in iCloud. Apple did not comply with the order and instead pulled its Advanced Data Protection service from UK users.

But within hours of the Washington Post story, Home Office sources confirmed the report to the Times. In October 2025, the Home Office issued a new TCN, reportedly narrower in scope and aimed at UK users.

As the tribunal was told, this was not the only confirmation of the original order. A post on X by former US director of national intelligence Tulsi Gabbard in August 2025 was seen by more than three million people, while in December 2025, the investigatory powers commissioner, Brian Leveson, said that “lawful access can be achieved in a way that strikes a balance between maintaining strong encryption and ensuring law enforcement and the government can protect the public from terrorism, serious crime, and hostile state activity.”

Gabbard had reported that the UK had agreed to “drop its mandate” for Apple to provide a “backdoor” that would have enabled access to the protected encrypted data of American citizens.

Jaffey told the court that it was unlikely that Home Office sources would have confirmed the TCN without “tacit approval” from the government. “But here, within hours of reporting by the Washington Post, multiple sources were confirming to the Times the existence of the orders,” he said.

Leveson’s statement, Jaffey continued, “gives the game away.” The commissioner’s report would have been “clearly prejudicial to issues that might come in front of him in the future” had there been no TCN issued at all.

And while the Home Office is sticking to its NCND policy, WhatsApp and Google have confirmed in witness statements that they have not received any such orders – something they are free to say because there is no legal bar on disclosing that you have NOT received a TCN.

Apple, on the other hand, is in a different position because it did receive the order, and is now in court trying to force the government to confirm this, and by extension, admit that the TCN is still in force.

Another argument made by Jaffey is that Apple had to ask for permission before it could discuss the order with the US government, and that permission was granted. “The fact that Apple even needed to ask permission confirms there is a TCN,” he said.

The government’s argument against abandoning the NCND policy is that it would damage national security and hinder the investigation and prosecution of “serious crime,” including terrorism and online child abuse. But Jaffey said that “the horse has long bolted” as far as secrecy around the TCN is concerned, and that the government’s stance is “ridiculous” and “understates the principle of open justice.”

Keep reading

UN moves forward with plans to take control of AI globally

Nobody panic. The UN and its vast globalist elite bureaucracy is ready to claim central oversight, control, and administration of future artificial intelligence (“AI”) development to ease worries about the fast-developing science.

Secretary-General Antonio Guterres made the declaration Wednesday, warning world leaders AI poses ‌risks – known and unknown – that cannot be ignored.

The former Portuguese Socialist Prime Minister spoke after pushback from President Donald Trump, who called threats that AI might destroy humanity “a HOAX” and claimed efforts to limit the technology are part of a “SICK conspiracy.”

He and Mark Zuckerberg both argue market pressure, not government mandates and centrally administered bureaucracy, will keep AI developers in line.

Guterres disagrees, embracing political and industry AI controls rather than trusting the marketplace.

Cautioning that “the world cannot afford a race to the bottom on AI safety,” the chief bureaucrat said concerns raised by those on the front lines of AI technology that its development “is racing ahead of our understanding of the risks” can’t be ignored, the Associated Press (“AP”) reports.

Governments must assume their first responsibility to protect civilians from AI threats, which means “national action is essential,” Guterres told UN reporters. “But global coordination is also indispensable.”

“We need guardrails to build trust – and that make AI safe, transparent, accountable, with human dignity at the centre.”

That’s where the UN and its strictures come in, Guterres outlined.

He said AI will be a major topic in his discussions with world leaders at their annual gathering next week for the high-level session of the UN General Assembly.

“The United Nations will continue to explore practical ideas and approaches in the immediate days ahead,” Guterres said, per the AP report.

“We need a shared understanding of how to advance the safe, secure and responsible development of AI – while identifying when increasingly powerful systems may require stronger safeguards, or additional measures to manage potential risks.”

Expect a global UN summit sooner rather than later on AI controls, Guterres said in response to a question asking whether next week’s meeting of top officials from the UN’s 193 member nations wouldn’t be the perfect opportunity for talks on curbing the excesses of AI.

The career bureaucrat said, “I think it makes full sense at the right moment to have a serious international summit on artificial intelligence.” And at a summit, he said, “it is absolutely essential to have everybody on board” – governments, companies, scientists and civil society – “and I hope that this will be possible.”

Keep reading

Senate Democrat Blocks Immediate Passage Of House-Passed Data Center Utility Rate Bill

A Senate Democrat on Sept. 17 blocked Republicans from moving to immediate passage on the Ratepayer Protection Act, a House-passed law that would encourage states to require data centers to pay for their own energy costs.

The bill previously passed the House in a 417-3 vote on Wednesday evening.

Though progressives in the lower chamber had expressed concerns about the bill’s lack of enforcement mechanisms, the faction ultimately backed the bill. It seemed on track for a painless passage of the Senate under unanimous consent, a procedure that allows uncontroversial bills to pass quickly.

The bill was introduced on the Senate floor Thursday by Sen. Jon Husted (R-Ohio), the bill’s sponsor in the upper chamber, who noted that as many as 1,000 new data center constructions are expected in the United States in the next five years.

“Americans are becoming ever more reliant on computing power from data centers for their daily lives, from medicine to manufacturing, education to agriculture, national defense, and the way we deliver basic services. Data centers are used not just for AI; they support almost every aspect of modern life, including online shopping, streaming video, social media, and the broadcast of this very debate.”

The law would require states to consider the adoption of models to require data centers to fund their own energy within a year, but would set no requirements that states ultimately adopt those models.

Husted moved for unanimous consent to pass the legislation and was blocked by Sen. Martin Heinrich (D-N.M.), who said that the bill to address this topic should have more stringent federal enforcement mechanisms.

“It’s not enough for us to tell states to consider making data centers pay for grid updates. Rather than voluntary pledges or suggestions to states, Congress needs to pass real legislation with real teeth,” Heinrich said.

Heinrich noted that for many Americans, utility rates had already increased – and that such Americans wouldn’t gain relief from the state opt-in process of the Ratepayer Protection Act.

Instead, he proposed consideration of his bill, the Grid Savings Act, “to require – not a voluntary agreement – to require large-load customers like data centers to pay for the facilities needed to connect them to the grid. Not to consider it, not to make pledges – to actually do it, to pay for the upgrades they need.”

Data centers are top of mind for many American voters this year.

The data centers used to power artificial intelligence (AI) services such as ChatGPT, Claude.ai, Grok, and others – aside from their extensive land requirements – require massive amounts of energy.

Estimates indicate that currently, data centers use about 4 percent of the U.S. electric grid, according to online energy shopping and consultation platform Electric Choice.

The energy used to power these facilities, which draw roughly 176 terawatt-hours of electricity per year, could otherwise be used to power 16 million homes, according to Electric Choice.

The Energy Department and other analysts have estimated that that share could skyrocket to 12 percent of the national power grid in the coming years.

Despite promises by many data centers to provide their own power, residents in areas where such centers have been built have reported spikes in their energy bills.

Anxiety over the possibility of such utility price jumps has driven many voters to be skeptical or outright hostile to the construction of such facilities in their area.

Likewise, cooling the computers that execute AI queries requires significant amounts of water, raising fears that energy might not be the only utility cost affected by the data centers.

Other residents living near data centers have also reported that the facilities are noisy. According to these residents, the massive fan and cooling system used to keep the servers from overheating is loud enough to travel through walls and disturb sleep.

Keep reading

Flock Camera Ping Jails Innocent Woman For 13 Days

A 23-year-old Palm Coast Florida woman spent 13 days in a Volusia County jail cell facing life in prison after Florida Highway Patrol troopers treated a Flock Safety camera hit like a confession.

Lindsey Brooke Isaacs did not cause the crash. Her black Dodge Durango had no collision damage. A 911 caller had already described a maroon Durango and a partial plate that was not hers. None of that stopped the machine-assisted hunt.

The October 4, 2025 hit-and-run on eastbound I-4 near DeBary killed motorcyclist Joaquin Deno, Flagler County Deputy Administrator Jorge Salinas, and Salinas’ wife, Nancy. A fourth person was badly hurt. Witnesses described a Dodge Durango fleeing the scene. Investigators searched nearby Flock cameras for a black Durango and found Isaacs’ tag, RJVN10, crossing the Seminole-Volusia line at 9:51 p.m., about three miles west of a wreck that started around 9:53 p.m.

That ping became the case.

Troopers later pounded on Isaacs’ door while she was asleep. “I thought I was getting set up. I didn’t understand what was going on,” she said. They seized the Durango for months and claimed “marks” consistent with the crash.

Photographs later showed an immaculate 2025 SUV. “Where is the damage on the vehicle? Because I’m standing right in front of it and I don’t see any damage,” Isaacs recalled.

On April 17, 2026, FHP booked her on eight felonies, including three counts of vehicular homicide and three counts of leaving the scene of a crash involving death. She was held without bond, then sat in G Block for 13 days. “Being incarcerated, falsely incarcerated, in jail, in G Block, was the worst 13 days in my life,” she said. “I was devastated, I was scared. I thought my life was over.”

Attorney Patrick McGeehan did the work the camera search replaced. A time-and-distance analysis using the same Flock timestamp put Isaacs past the crash site when the collision happened. “In Lindsey’s case, there was nothing indicating that she was ever there,” he said.

A 911 caller described a maroon Durango and the partial plate “458.” Paint transfer on a struck Ford Focus was maroon. Isaacs’ plate did not match. Witnesses had described three different vehicles: black, white, and maroon. “For some reason, [Florida Highway Patrol] never went past the black Dodge Durango,” McGeehan said. “They hooked onto this black Dodge Durango to the exclusion of everything else.” He called it “confirmation bias, to the highest degree” and “a major failure of basic investigative skills.”

“Lindsey wasn’t even at the crash scene,” he told NewsNation. “These were investigators that got it in their mind that there was a black Dodge Durango involved. They got on Flock, found the first black Dodge Durango, that belonged to Lindsey Isaacs, and pinned that on her.”

Keep reading

Researchers Used Claude To Hack OpenAI Employee Accounts

Three security researchers used Anthropic’s Claude to breach OpenAI employee accounts and gain access to private company software in a July attack that began with an image upload to the company’s public help forum.

The researchers, Harsh Jaiswal, Mohan Pedhapati and Rahul Maini of Hacktron AI, described the July 25 breach in a report published September 13. They said the work took less than 72 hours from the initial discovery to demonstrating access to an internal OpenAI software repository. OpenAI subsequently paid them a $6,500 bounty.

The disclosure follows a separate incident earlier in July in which OpenAI’s own AI agents escaped a testing environment and attacked Hugging Face, a platform used to host AI models and datasets. In that case, OpenAI says the agents took dangerous actions that weren’t directed by a human – the incident being used to spook everyone into letting far-left technocommies run AI oversight. 

Hacktron’s team directed its own operation, reported the vulnerabilities to OpenAI and stopped after demonstrating access.

How A Forum Upload Reached Internal Software

The entry point was OpenAI’s public discussion forum, which runs on software supplied by Discourse (3rd party software that manages discussion boards). A flaw in an image-processing component called libheif allowed a specially crafted image upload to make the server execute the researchers’ instructions. Discourse’s security advisory confirms the vulnerability required no interaction from a victim.

Hacktron said the faulty code had been changed in 2025, but the change was not identified as a security fix. The forum was still running a vulnerable version.

Keep reading

WW3 Near-Miss? AI Hallucinated Nuclear Weapons Components Aboard Chinese Vessel Bound For Iran

Could runaway artificial intelligence (AI) spark a future WW3?

We don’t need movies like Terminator 2 and its SkyNet to help us imagine this, it’s 2026 and there is already a terrifying real-world precedent that could have kicked off a shooting war with nuclear-armed China.

A human analyst within the US intelligence community utilized data gathered by AI for a report that was later widely circulated across classified Pentagon channels, except that the chatbot which was relied upon in effect hallucinated nuclear weapons components being transported aboard a Chinese vessel bound for Iran.

Naturally this set off the highest alerts in Washington in the middle of an active hot war which President Trump launched ostensibly to prevent this very thing – Iran obtaining a nuke and atomic weapons components.

From there military planning kicked into high gear, with a special forces boarding operation imminent. Supposedly elite assets were already airborne, ready to descend on the Chinese ship in question.

But then, “the operation was halted only after officials took a closer look at the intelligence report and discovered that it had been produced with assistance from an AI chatbot,” CNN reports in a bombshell Friday exclusive. The episode reportedly happened last spring.

The severe error was realized with barely enough time to call off what could have been the geopolitical blunder and disaster of the century, “The chatbot had incorrectly identified the material the vessel was carrying.“

In essence the United States found itself based on completely manufactured ‘intel’ potentially headed for conflict with China.

As if to to preempt anyone who might be tempted to think the account is merely being hyperbolic or needlessly sensational, one official (unnamed) source told CNN that the intelligence assessment, which was “entirely false” – in the end “almost started a war.“

While Pentagon leadership, including US Special Operations Command Pacific, has yet to respond to CNN’s queries, the outlet pieces together the alleged chain of events and how this could have possibly happened in the following:

In this particular instance, the analyst queried a chatbot about some intelligence reporting on the ship’s manifest that originated with US Special Operations Command Pacific, based in Hawaii. It was not clear whether the chatbot was a commercially available one or a US government product.

“The internal tools are mostly just copies of the commercial stuff wearing lipstick,” a former senior US official familiar with the AI systems used by military and intelligence analysts.

The bot fused together open-source intelligence with secret signals intelligence in government holdings and reached its fateful conclusion about the material the ship was carrying.

The analyst then used AI again to package the findings into a standard intelligence report — the kind that is trusted by military officials — and disseminated it.

The report seems to be suggesting that the unknown AI tool that was utilized would be deemed substandard and weak for such an important assessment in the first place.

Further “comforting” for readers is that given heavy AI-integration seems the “irreversible” trend of the day across the DoD/Depart of War and IC, similar near-misses in the near and distant future are bound to happen.

CNN underscores, “Across the US military and the intelligence community, officials are pushing to weave AI into nearly every facet of their work, from analyzing the huge volumes of raw intelligence the US collects and selecting targets for strikes, to more mundane applications like managing budgeting, logistics and supply chains.”

Keep reading

Newsom Signs Executive Order Directing California to Develop an AI Kill Switch Plan

California Gov. Gavin Newsom signed an executive order Friday directing California agencies to speed up independent oversight of frontier AI companies and study an emergency shutoff, a “kill switch,” for the most powerful models.

The order tells the Government Operations Agency, working with the Governor’s Office of Emergency Services, to work with experts and deliver recommendations within two months.

Newsom said in a statement:

The federal government’s abject failure to create any form of meaningful AI oversight or accountability should alarm every American, especially when AI CEOs themselves are begging for regulation.

We’re not waiting to act – we’re going to speed up our work on substantial and responsible AI oversight before it’s too late. We’re going to do this thoughtfully but with urgent velocity; the stakes are too high to wait or delay action.

While Washington abdicates its responsibility to protect Americans, California is building on the strongest AI regulatory framework in the nation. California has already built a national model, and our policy should be the national baseline.

Keep reading

Ukrainian Drone Hits Russian Nuclear Plant Days After Trump Energy Truce Plea

President Trump raised eyebrows when on Monday he declared that Russia and Ukraine had agreed to stop attacking each other’s oil and energy infrastructure, amid global crude and diesel supply problems related mostly to the Hormuz Strait crisis.

But then within days later it became clear that the default reality on the ground has persisted, as energy sites continue to get pummeled. And it’s not only refineries and ports which have continued getting hit, but as of Thursday morning a Russian nuclear plant was apparently targeted in a Ukrainian drone attack.

A drone hit the cooling tower of a reactor unit at Russia’s Kursk Nuclear Power Plant, the International Atomic Energy Agency has announced Friday.

The unit was operating at the time, but its operating mode was unchanged and no fire broke out. 

The UN agency’s director Rafael Grossi declared in a statement that “all attacks on nuclear facilities are unacceptable as they could endanger nuclear safety and security, irrespective of where they occur.”

He urged “maximum military restraint to prevent the risk of a nuclear accident” – which remains a top dangerous reality of the battlefield in the context of the Ukraine war.

In early July the same site had suffered a prior drone attack from Ukraine. One industry source reviews:

The Kursk Nuclear Power Plant (Kursk NPP) has been the target of several military actions and drone strikes. The most notable incident occurred in August 2025 when a drone detonated and damaged an auxiliary transformer, sparking a structural fire that had to be contained by emergency teams. The International Atomic Energy Agency (IAEA) monitored the event and confirmed that radiation levels remained normal.

The Kremlin going back to 2024 has accused Ukrainian forces of seeking a ‘provocation’ at the Kursk nuclear plant.

The Kursk Nuclear Power Plant (NPP) lies about 25 miles west of the city of Kursk, which is the administrative center of the oblast by the same name.

The plant continues to be a major electricity producer for Russia. Kremlin officials have long voiced alarm that Ukraine may seek to use some major provocation to draw its backers in NATO deeper into the conflict.

Keep reading