Tag: ai

OpenAI says its AI agents disabled their own safety monitors

OpenAI has revealed that two of its artificial intelligence agents disabled or weakened systems built to catch them misbehaving, with one later denying it had done so.

The cases are the latest in a string of safety scares at the world’s top AI developers involving agents, AI programs that carry out tasks without human involvement.

In a report released Tuesday, the ChatGPT maker said that during two internal coding jobs, its agents switched off or undermined tools for detecting incidents of AI models acting against human wishes.

On October 3, an agent running on OpenAI’s GPT-6.1 Sol model, which had been told not to send any messages on the workplace chat app, took that as license to switch off safety alerts meant for the company’s security team while fixing a testing system.

Three days later, a GPT-6 Astra agent stripped out its own monitor while copying code from another project, then told the human programmer it had done no such thing.

The company said it had also seen several cases of agents launching software with monitoring switched off, exposing a gap in its oversight.

OpenAI said it is still trying to determine whether the models were simply overzealous in following instructions or have a particular tendency to evade monitoring.

“We don’t yet have a satisfying explanation for this behavior,” it said.

The disclosure adds to mounting unease over how well leading labs can control increasingly autonomous systems.

In July, a group of OpenAI agents broke out of their test environments and launched a cyberattack on AI firm Hugging Face, causing international alarm. Other companies have since reported similar incidents.

The incidents have fueled calls in the United States and abroad for tighter oversight of AI development.

But US President Donald Trump, under the advice of top AI executives like Nvidia chief Jensen Huang, has urged the industry to police itself.

Keep reading

Pentagon Letter Lays Out Plan For Four-Star Autonomous Warfare Command

On October 19, the War Department put out a Pentagon letter on how they plan to build a four-star Autonomous Warfare Command. The letter, on Project Agincourt letterhead and signed twice by officials marked “CEO,” began by referencing the past attempt by the Army to transform itself in 1928.

The letter is titled “Building the Autonomous Warfare Command” and comes from the Direct Reporting Portfolio Manager for Unmanned Systems (DRPM-UxS), which leads Project Agincourt. The letter revisits a historical episode from the interwar years of the War Department’s attempt to build a unified mechanized force of tanks, armored trucks and motorized infantry. In 1928, Secretary of War Dwight Davis directed the Army to experiment with such a force. However, that force was soon disbanded, and the effort at mechanization was ultimately broken up and added as an afterthought to the various services. It required the German blitzkrieg in 1940 to force the creation of the U.S. Armored Force.

“We face an equally urgent revolution,” the letter says, pointing to low-cost drones that have changed the character, speed and economics of war. Instead of going down the path of 1928, it says, Secretary Pete Hegseth “chose the maximal path.”

Project Agincourt has two missions. The first is to stand up the Autonomous Warfare Command (AUTOWARCOM) by October 1, 2027, as the largest institutional change since the interwar years. The second mission of Project Agincourt is to continue the mission of the DRPM-UxS while pioneering a new acquisition model, which pairs autonomy and robotics companies with operational forces at the “edge.” “The goal is not a particular drone. It is a system built to adapt at scale,” the letter states.

Power projection since the Cold War has been based upon a limited number of very high-end, very expensive systems, and that model has been disrupted. The new advantage for the U.S. will come from the edge, adapting faster than adversaries with two key assets: its lead in superintelligence and an unmatched NCO corps.

Keep reading

Lawmaker Seeks Details on Secret Data-Center NDAs as AI Build-Out Faces Scrutiny

Rep. Jamie Raskin, the top Democrat on the House Judiciary Committee, sent letters Tuesday, Sept.29, to Amazon, Google parent Alphabet, Meta Platforms, and Oracle seeking information on non-disclosure agreements (NDAs) signed with government officials related to data center build-outs, according to copies viewed by The Wall Street Journal.

Raskin wrote that residents cannot weigh a subject’s costs against its benefits, hold officials accountable, or safeguard resources when basic terms are kept secret, according to the report. He demanded detailed information on each company’s data center portfolio and use of the agreements [7].

Amazon claims it no longer uses NDAs with government collaborators on data center projects, according to the report. Google, Meta, and Oracle did not immediately comment [7].

The letters mark an escalation in congressional scrutiny of an industry that has expanded rapidly across the country. The requests focus specifically on agreements that shield the terms of data center deals from public view [7].

Local and State Officials Cite Secrecy in Opposition

Local communities have cited non-disclosure agreements and secrecy surrounding data center plans as a major reason they oppose the facilities, which many blame for pushing up power prices and consuming large amounts of water [7].

Data centers are being sited across the United States without meaningful public input, a pattern documented in states such as Utah, where the three-member Box Elder County Commission voted unanimously to greenlight a 40,000-acre hyperscale project before residents could weigh in [1].

The backlash has become a measurable force. Local opposition disrupted at least 45 proposed U.S. data center developments worth nearly $68 billion between April and June 2026, according to Data Center Watch, a research project operated by AI intelligence company 10a Labs. The sustained resistance now spans 49 states with 843 active opposition groups, according to the same research.

Data centers have become a flashpoint in key races including Ohio, Michigan, and Texas, where states have taken steps to cut tax exemptions worth billions of dollars. A new Wall Street Journal poll found that 62% of voters support pausing data-center development, according to the report.

Pennsylvania Gov. Josh Shapiro banned the agreements in an executive order in August, and several states have introduced bills to ban data-center NDAs. In Texas, Gov. Greg Abbott ordered the state’s environmental watchdog to withhold permits for new data centers until an audit of risks to the power grid is complete, according to ZeroHedge [2].

Keep reading

Situational Unawareness: Banks Ran Out Of Room To Lever Up The AI Trade, So They Made More

When a bank runs out of room to lend to hedge funds piling into the most crowded trade on the planet, there are two options: stop lending, or find more room. Barclays found more room…

According to a Bloomberg exclusive published Tuesday, Barclays (BARC) executives wanted extra capacity to lend more aggressively to hedge fund clients as Asian AI stocks went vertical earlier this year. The fix: the bank’s macro rates traders ended up lending part of their own capital allocation to the prime financing business through an internal auction. The reason they had to: over the summer Barclays had reached its maximum limits on how much financing it could extend to hedge fund clients in Asia. Bloomberg’s verdict: Barclays and its peers are “finding ways to meet demand, despite bumping up against internal constraints.”

In other words, the internal risk limits did their job, flagged that the bank had lent enough… and the bank responded by moving the goalposts.

“Disciplined”… Right Up To The Limit

The pitch inside Barclays, per Bloomberg’s sources, was simple: allocate more balance sheet to higher-returning businesses like prime brokerage (housed in the global markets unit run by Adeel Khan) given the “huge growth” expected in Asian tech stocks. Rates traders, whose returns on balance sheet don’t look nearly as sexy as financing Korean memory stocks at peak euphoria, became the donors.

Barclays’ official line:

“We deploy our investment bank balance sheet dynamically, responding to client opportunities while maintaining a disciplined approach to risk and returns.”

Translation: we found a client opportunity, and we deployed dynamically. Meanwhile, the “disciplined” part kicked in only after the limit was hit.

Regular readers will recall that the Bank of England had already started poking around exactly this corner of the market just as the AI trade was cracking in July.

And as we noted when regulators finally found their situational awareness (Sep 24), the BoE is specifically looking at the rapid growth of London prime brokers financing Asian equities this year, after names like SK Hynix went vertical. London prime brokers financing Asian equities… it’s almost as if someone had a specific bank in mind.

Keep reading

Axios Uses Fake AI Image to Run Damage Control for Somali Politician Who Spoke Next to Upside-Down Flag

To defend a leftist politician who spoke in front of an upside-down American flag, Axios published a fake fact check claiming the video was “doctored” by AI and that, in reality, the flag was right side up.

Axios just retracted that fact check.

Naturally, Axios would like us to believe its fake fact check was an honest mistake, but Normal People know two things about dishonest, leftist outlets like Axios: 1) their “mistakes” only ever fall one way, and that’s to benefit the left, and 2) Axios was in a panic to defend a leftist, so they rushed to publish this — something Axios would never do for a Republican.

Oh, and 3) fact-checking is a hoax, a leftist psyop, a disgraced profession that has spewed more lies than Joe Isuzu.

Side note: Fai is currently a Washington County Commissioner. Officially, she holds a non-partisan office, but she is a leftist.

On October 10, Washington County, Oregon, Commissioner Nafisa Fai spoke at a Somali event in Minneapolis. A few days later, the End Wokeness X account published a portion of that speech that clearly showed her speaking with an upside-down American flag behind her.

“Oregon politician Nafisa Fai just held a rally in MN, speaking Somali, with an American flag UPSIDE DOWN,” wrote End Wokeness. “Enough with the pretending already. These are not Americans.”

Keep reading

Meta Muse Exposes Users’ Personal, Business Secrets When They Share Conversations

Security researchers say Meta’s Muse AI assistant exposed thousands of users’ shared conversations because conversation titles were placed in subdomains that could be discovered through public DNS records. Meta has since suspended the sharing feature while it considers making changes. Ars Technica reports:Like many online services, Muse has allowed users to share searches, conversations, images, and similar content with other people. Such share features usually involve clicking an option to create a unique URL that anyone can use to load a page hosting the content. To keep potentially sensitive information from being accessible to unintended people, the URLs typically contain a long sequence of text that can’t be easily guessed.

Muse, which Meta CEO Mark Zuckerberg said last month was “built from the ground up for privacy and security,” seems to take the same precautions. But a central design decision for constructing URLs has made the links easy to discover. As a result, researchers from security firm Claroc said they were able to enumerate 6,800 links Muse users created for sharing conversations with others. After opening a few hundred of them, the researchers found information that was clearly not intended to be publicly available… […]. “If you had asked us beforehand, we would have guessed that gathering these would take real effort,” the researchers wrote. “It took one query.”

[…] The easiest way to discover the URLs is to perform a passive DNS search through one of many available commercial services. Passive DNS records store the historical resolution of domain names to IP addresses, as well as nameserver, and mail-server records over time. Passive DNS systems typically collect DNS responses from sensors positioned at recursive DNS resolvers or other infrastructure. A passive DNS record is generally created when a monitoring system observes a DNS response. In the case here, once the intended party clicks on the share link, the DNS information is recorded in a searchable database.

Claroc researchers were able to discover the Muse share links because all of them use the domain name cf.metaaiusercontent.com. Discovery is further made easy by the decision to add the conversation name to the URL, along with a long string of text. Simplified examples of such links are home-sale-budget-9f21c8a3.cf.metaaiusercontent.com, q3-revenue-review-7b1e42f0.cf.metaaiusercontent.com, and loan-application-draft-a4c9e2.cf.metaaiusercontent.com. It’s this last design decision, putting the conversation title in the subdomain rather than in the URL path, that makes public discovery possible.

Keep reading

Brookfield, NextEra To Build $100 Billion AI Hub On Cold War Nuclear Site

The U.S. Department of Energy (DoE) has announced a landmark $100-billion private investment to turn the former uranium enrichment site in Western Kentucky into the new Paducah American Energy Hub, marking one of the largest economic development projects in Kentucky’s history.

Led by U.S. Secretary of Energy Chris Wright, the initiative will bring together a private-sector coalition to build a massive artificial intelligence and high-performance computing (HPC) innovation campus powered by new energy infrastructure. The phased buildout is targeted for 2032, and is expected to create roughly 8,000 construction jobs and 600 permanent operations jobs. Brookfield Asset Management (NYSE:BAM) has been selected to build the 1.8-gigawatt (GW) AI and HPC data center campus while NextEra Energy (NYSE:NEE) will develop 2 GW of natural gas-fired generation coupled with 2.6 GW of battery energy storage. The project is designed to operate on dedicated power supplies capable of feeding excess energy back to the regional grid to avoid raising utility bills for local families and businesses. The campus will also deploy a modern closed-loop cooling system to continuously recirculate water and minimize external consumption. The project leaders emphasize that this design is vastly more resource-efficient compared to the site’s previous tenant, the Cold War-era Paducah Gaseous Diffusion Plant, which was originally designed to pull up to 30 million gallons of water per day from the Ohio River.

Still, the Paducah project comes at a time when the market is souring on heavy AI infrastructure spending.

Last year, Trump, alongside Oracle’s Larry Ellison, OpenAI CEO Sam Altman and SoftBank CEO Masayoshi Son, unveiled Stargate, a private-sector joint venture intended to invest up to $500 billion over four years to build hyper-scale AI data centers and energy infrastructure across the United States. Billed by Trump as the “largest AI infrastructure project in history,” Stargate is designed to secure American dominance in AI over foreign competitors like China. Oracle serves as the core cloud and infrastructure provider for the venture. Initial development of the project began in Abilene, Texas, consisting of multiple 500,000-square-foot data center buildings to house the immense computing power needed for next-generation AI models.

Unfortunately, the ambitious AI infrastructure project found itself on the ropes just months after its grand launch.

Back in January, Business Insider reported that JPMorgan Chase was having trouble syndicating the $38 billion in construction debt it underwrote for the Stargate AI data center initiative as institutional investors grow increasingly cautious regarding concentrated tech sector exposure. Banks spent more than six months attempting to distribute the multi-billion-dollar debt packages for the Stargate campuses located in Texas and Wisconsin; however, buyers appeared uninterested in taking on massive piles of debt tied heavily to single-tenant AI operations. Oracle leases the facilities, and S&P cut Oracle in July to BBB-, one notch above junk, over its data center spending. Investors demanded a higher return to hold the debt, and the banks had to sell it at a discount.

In April, OpenAI officially paused its Stargate UK data center project, citing high industrial electricity prices and regulatory uncertainty. Stargate UK was planned in partnership with hardware firms Nvidia (NASDAQ:NVDA) and Nscale, aiming to deploy thousands of advanced AI chips (starting with 8,000 GPUs and scaling up to 31,000) across sites in North-East England, including Cobalt Park near Newcastle and Blyth.

Whereas OpenAI did not cancel the project outright, the AI giant says it will only resume development only when the “right conditions” allow for long-term infrastructure investment. The pause dealt a major blow to UK infrastructure partner Nscale (founded by Josh Payne), which had been scaling up operations and weighing a potential stock market IPO, as well as Britain’s push to emerge as a sovereign AI superpower.

The UK pause is the second blow to OpenAI’s planned capacity in a month.

Last month, Bloomberg reported that Oracle’s pivotal $18-billion Project Jupiter, a massive 1,400-acre data center campus in New Mexico critical to supporting the broader AI infrastructure goals, has stalled. Oracle issued a force majeure notice to lender Blue Owl Capital regarding the 2.45-gigawatt hyperscale data center in southern New Mexico after the tech giant failed to secure vital state and federal land-use permits for a planned natural gas plant. Instead, Oracle announced plans to shift over to solid oxide fuel cells (SOFCs), an experimental stopgap that has never been deployed on a scale this large.

Public resistance to data centers has been growing rapidly across the United States, Europe and Asia as communities push back against the massive environmental and economic footprint of AI infrastructure. A recent Gallup poll shows that roughly 71% of Americans oppose the construction of AI data centers in their localities. Pushback in the U.S. has resulted in projects worth over $130 billion blocked or delayed, with reports indicating more than 500 active data center bans or moratoriums have popped up nationwide.

Keep reading

Trump Admin, US Businesses ‘Need To Be Wary’ Of AI Espionage, Intelligence Chief Warns

Director of National Intelligence Jay Clayton warned that both the Trump administration and American companies “need to be wary” of foreign espionage with the rise of artificial intelligence (AI), which the government now refers to as super intelligence.

Clayton, who leads the Trump administration’s Super Intelligence Force, told CNBC in a Friday interview that maintaining U.S. leadership in the AI race is critical in the face of potential foreign espionage. He noted that while AI platforms and capabilities can be used for “very good purposes,” they can also be used for “nefarious purposes.”

The intelligence chief’s comments came after President Donald Trump’s Super Intelligence Force held its first meeting at the White House on Thursday. Clayton told CNBC that the meeting was “very good” and that it built on the voluntary accord signed by Trump and multiple technology industry leaders at the White House on Sept. 29.

Asked whether Americans should be concerned about foreign espionage with AI companies, Clayton said Friday that both commercial and state-sponsored espionage are a reality facing the United States.

“It’s something that our private-sector companies need to be wary of, we as a government need to be wary of,” he said.

Clayton told CNBC that it was important for the United States to “stay on the cutting edge” of AI capabilities in order to address concerns regarding AI advances by U.S. adversaries.

“We’re not the only ones developing it. There are vectors in this technology that are coming from some of our known adversaries and competitors,” he said. “The Chinese, the Russians, the Iranians, the North Koreans all have significant technical capability. We need to stay ahead of them.“

Many of the risks posed by AI are linked to cybersecurity and digital infrastructure, Clayton told CNBC. He described the U.S. digital infrastructure as critical infrastructure, covering everything from public utilities, such as water and electricity, to national defense.

While cybersecurity risks are not new, the rise of AI “amplifies” those risks, Clayton said. To address the risks, he said the United States is “focused on hardening those systems” and is committed to staying ahead of “nefarious actors” in the AI race.

Clayton’s warning regarding foreign espionage attempts linked to AI came after prosecutors and company officials said on Oct. 13 that Abdulrahman Alfozan, a Saudi national and software engineer working for OpenAI, was arrested on allegations that he concealed a sophisticated recording device in an OpenAI office.

According to a court complaint, the office Alfozan allegedly planted a recording device in was frequently used to hold meetings involving highly confidential information, including government engagements, commercial relationships, product plans, and strategic decisions.

In a Tuesday press release, Clayton described the incident as a “serious situation [that] underscores why winning the global super intelligence race is a national security prerogative.”

During Friday’s interview on CNBC, Clayton added that the incident involving the Saudi software engineer at OpenAI “shows the lengths that people will go to try and supplant our leadership in this, or at least stay close to us, in this emerging technology.”

The intelligence chief’s warning also followed a series of AI swarm breaches, which has led some industry leaders to suggest that the pace of AI development needs to be slowed down and the technology needs additional oversight.

A report released last month by Anthropic found that the Iranian regime used the company’s AI models to track and target U.S. Navy ships in the Middle East. The report also said that a Chinese-linked operation used Claude to track and attempt to recruit Uyghurs who joined the Syrian army to become spies.

In addition, Anthropic’s report found that individuals affiliated with the Chinese regime used Claude and supporting Chinese AI systems to surveil groups targeted by the regime. Those targeted by the China-linked individuals included members of the Tibetan Buddhist civil society, practitioners of Falun Gong, Catholic cardinals, Presbyterian Church leaders, the New York-based Shen Yun Performing Arts, and NTD, The Epoch Times’ sister media outlet.

Trump announced the formation of the Super Intelligence Force earlier this month to oversee U.S. AI policies. Clayton told CNBC that the White House task force took a number of actions during its inaugural meeting on Thursday.

Those actions included outlining a plan to implement the accords signed by technology leaders, tasking the President’s Council of Advisors on Science and Technology with conducting a risk study, establishing committees covering different AI issues, and developing a system for the Department of Labor to regularly report on the impacts of AI on the U.S. workforce and economy.

Keep reading

In Virginia, Racism and Redlining Are Fueling a Flock Camera Backlash

Local governments in Virginia are under mounting pressure to cancel contracts for surveillance cameras known as automated license plate readers (ALPRs). On October 6, the Henrico County Police Division announced that it would terminate its contract for such cameras after residents shared “legitimate concerns about Flock Safety.” In nearby Richmond, Mayor Danny Avula said on October 13 that the city would cancel its contract with Flock Safety, the most famous purveyor of the cameras, but did not rule out using ALPRs from a different company.

“It is a good first step, but we do need to keep the pressure on the city to not automatically switch to another ALPR vendor,” Richmond Democratic Socialists of America (DSA) member Victoria McCullough told Truthout.

DSA activists and residents in Richmond pushing police to ditch the contract have powerful research to point to from elsewhere in the state. In the neighboring Hampton Roads region of Virginia, race and class are the primary predictors of locations where police installed ALPRs, a groundbreaking study found earlier this year.

The study found police were more likely to install ALPRs in majority Black census tracts in each of the nine cities across the Hampton Roads region, an East Coast metroplex with 1.8 million residents and multiple cities, including Norfolk and Virginia Beach. About 80 percent of majority Black neighborhoods contain at least one camera compared to fewer than 40 percent of majority white tracts. ALPR surveillance is also more intense in majority Black neighborhoods, where on average there are four times as many cameras than in white neighborhoods when weighted for population share.

In Virginia Beach, police deployed ALPRs in only five of 76 majority white census tracts (6.6 percent), compared to one of two Black census tracts (50 percent). Black residents of the locality are 7.6 times more likely than white residents to live in a census tract with at least one ALPR. When weighted for the number of cameras deployed in each area, Black drivers in Virginia Beach were roughly eight times more likely than white drivers to be regularly surveilled by Flock-style cameras.

Keep reading

BTIG Spots “Real-World Evidence” Humanoid Robots Are Entering The Workforce

Goldman Sachs’ recent upward revisions to global humanoid robot delivery forecasts through the end of the decade, along with rising estimates across Wall Street desks, point to broader expected adoption in factories, restaurants, and other workplaces. Military applications and household use could follow soon.

BTIG strategic opportunities analyst Jesse Sobelson told clients Friday that the physical AI boom is moving onto factory floors, into warehouses, and behind restaurant counters.

Sobelson and his team attended RoboStrategy’s Robotics Founders Forum in New York this week and reported back on what executives from several of RoboStrategy’s private-company holdings (Apptronik, Standard Bots, Dexmate, Maven Robotics, Path Robotics and Dyna Robotics) discussed: robot development, commercial deployments, and manufacturing plans.

Keep reading