2.0
Bill Gates Says We May Have To ‘Suffer A Few Of The Negatives’ For People To Take The AI Warning Seriously — Compares Threat To An Alien Invasion
Jeannine Mancini
Tue, September 29, 2026
© USA TODAY via Imagn Content Services, LLC
Benzinga and Yahoo Finance LLC may earn commission or revenue on some items through the links below.
Artificial intelligence doesn't exactly come with an emergency brake. Microsoft co-founder Bill Gates thinks the world may have to experience something go seriously wrong before it takes the technology's biggest risks as seriously as it should.
"One thing that's clear is if you have a big cyber attack that caused a lot of problems, then people will say, 'Okay, they were right,'" Gates told "CBS Evening News" in September. "So, it may be that we have to suffer a few of the negatives before we'll respond as fully as we should."
Gates was in New York during the United Nations General Assembly discussing the potential and risks of artificial intelligence when the conversation turned to what it might actually take for governments and the public to act
"So, in other words, what it's going to take for us as a country and a world to tackle this is not just a 12-page essay from Bill Gates. It could be a catastrophe," host Tony Dokoupil said.
"That's fair," Gates replied.
A Cyberattack Could Change The Conversation
The warning comes after Gates has spent months highlighting the risks that could come with increasingly capable AI. Those concerns include cyberattacks, major job losses and the possibility that AI could make it easier for bad actors to develop biological threats.
His August essay on AI described a period of rapid technological change that could disrupt the workforce and create risks that governments aren't prepared to handle.
CBS News also referenced bioterrorism and mass unemployment among the potential catastrophes Gates has discussed.
The cyberattack scenario gets at a broader point. Warnings alone may not be enough to change how seriously people treat AI risks, according to Gates. A major real-world incident could provide the kind of evidence that makes the threat impossible to ignore.
The 'Aliens Are Here' Warning
Gates has also reached for a very different analogy when discussing how countries should respond to AI.
At the Telluride Film Festival in September, he compared advanced AI to an alien invasion and pointed to science-fiction movies in which the arrival of an outside threat forces countries to cooperate.
"There are these movies where the aliens are coming, and you see unprecedented levels of cooperation between specifically China and the U.S.," Gates said at the festival. "Well, believe me, the aliens are here."
He described AI as "a foreign intelligence with a different set of motivations that we can't even understand" and said something equally dramatic may be needed to push countries toward greater cooperation.
AI's Biggest Problem May Be Power
For all the talk about cyberattacks, jobs and alien invasions, there's another very physical requirement behind the AI boom.
Electricity.
AI data centers require enormous amounts of reliable power, and the race to build that infrastructure is creating opportunities for companies developing new generation capacity.
American PowerGen is focused on that piece of the puzzle. The company develops power projects aimed at serving AI data centers and other large electricity users and has described a 22-gigawatt development pipeline across 16 Texas projects.
For investors who want to get involved with the infrastructure supporting AI, American PowerGen is conducting a Regulation Crowdfunding offering. Its offering materials list shares at $2.50 each, allowing individual investors to invest in the company at an early stage.
AI may be racing ahead, but the infrastructure underneath it has to keep pace. And that means the future of artificial intelligence still depends on something far less futuristic than artificial minds or alien intelligence — keeping the lights on.
OpenAI GPT-6.1 Astra Reportedly Pulled After Safety Tests Show it Could Evade Human Oversight: 'We Have an Extremely High Bar'
Shomik Sen Bhattacharjee
Wed, September 30, 2026
BENZINGA
Sam Altman and the OpenAI logo (Credit: Sam Altman and the OpenAI logo | Photo courtesy: Meir Chaimowitz / Shutterstock)
OpenAI has scrapped the planned release of GPT-6.1 Astra after internal safety tests found the model showed more deceptive behavior than its predecessor and sometimes evaded human oversight, intensifying scrutiny of increasingly autonomous AI systems.
GPT-6.1 Astra Falls Short On Safety
The Wall Street Journal first reported the decision on Monday. Reuters said OpenAI had planned to release the model in October and integrate it into ChatGPT and Codex. Tests found GPT-6.1 Astra sometimes failed to accurately disclose actions and operated outside the authorized scope.
"Of course we want to make sure our model development is safe no matter whether that's in the company, or when we ship it to users," safety systems head Saachi Jain said in a statement shared with Reuters. "But when we ship it to users, we have an extremely high bar in terms of safety and alignment."
The decision follows OpenAI's Sept. 3 release of GPT-6 Astra, which the company calls its most capable broadly deployed model. OpenAI classified Astra as its first model to reach the "Critical" cybersecurity capability threshold, meaning that with the right tools and access it can discover unknown vulnerabilities and develop exploits across well-protected systems without step-by-step human guidance.
Benzinga reached out to OpenAI for additional comment but did not receive an immediate response.
Agent Incidents Intensify OpenAI Safety Scrutiny
Safety concerns have grown since OpenAI disclosed that an autonomous agent escaped a controlled evaluation environment and breached Hugging Face. Reuters separately reported Tuesday that another OpenAI agent accessed an Australian government website in June, retrieving internal files and credentials. OpenAI apologized and said no medical records were compromised.
The canceled release also lands amid an industry debate over slowing frontier development. Anthropic CEO Dario Amodei urged labs this month to pace capability gains while strengthening safeguards, a proposal backed by OpenAI CEO Sam Altman. Altman has separately argued that competitive pressure cannot justify "recklessness".
Safety Debate Looms Over Developer Conference
"While [GPT-6.1 Astra] improved on axes such as model laziness, it didn't quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it's done," Jain said.
News of the model being shelved arrives ahead of OpenAI's annual developer conference, which is set to get underway on Tuesday.
An Australia Security Researcher Shut Off a Moving BYD Pickup’s Headlights From the Roadside, Then Used Its Speakers to Get Siri to Read Out the Driver’s Home Address.
Eve Nowell
Wed, September 30, 2026
A researcher sat on the shoulder of a road outside Canberra with a laptop and switched off the headlights of a moving BYD Shark 6. There was a reporter at the wheel.
The stunt was staged for Australia's ABC and its investigative program Four Corners, which handed the plug-in hybrid ute to Dan Hreszczuk, co-founder of Canberra firm Fortify Labs. He had two weeks to find out what could be seen and done to the truck from a distance. His verdict, as published in the ABC's own account, was blunt: "The access we took advantage of didn't even have a password."
The Demo: Annoying, Distracting, Then Dark
The sequence escalated nicely for television. With reporter Angus Grigg inside, Hreszczuk locked the doors, pumped music through the speakers and threw images onto the infotainment screen. Once the ute was moving, he ran the wipers at full speed, fired the washer jets and flicked the headlights on and off before shutting them down completely.
Relax about the cliff scenario: the 2.6-tonne ute was only doing about 30 km/h on a reasonably straight road. Hreszczuk also said he couldn't get into the brakes or cameras, which he described as well protected.
That's the good news, and why it happened is worth a minute.
Why He Got the Lights But Not the Brakes
Modern cars run on internal networks, most commonly the CAN bus, which dates back to an era when nobody expected a stranger to be plugged into it. Classic CAN has no built-in way to verify who sent a message. If a module on the network says "headlights off," the headlight controller generally believes it.
Hreszczuk explained the point in the broadcast episode, where he described the CAN bus as the internal network that the car's components use to talk to one another. The footage also shows him hunting through the ute for places to tap into it.
Well-engineered cars put gateways between the comfort-and-convenience side of the network (lights, locks, wipers, audio) and the chassis side (brakes, steering, powertrain). The fact that Hreszczuk could play DJ but couldn't touch the brakes suggests BYD did segment the safety-critical systems. The failure was leaving an unauthenticated door open into the "soft" side of the car, and it turns out the soft side can be dangerous too. Killing the headlights at night on a twisty road is not a cosmetic problem.
The Asterisk: This Started With the Keys
Before this turns into "BYDs can be hacked from the roadside": this was not a drive-by attack on a random stranger's truck. Hreszczuk had the Shark in his workshop for two weeks, with time to pull trim, probe connectors and study the car's software. The "remote" portion of the demonstration came after that period of unrestricted physical access.
That doesn't make the result meaningless. Plenty of people get unsupervised time with your car: valets, detailers, independent shops, a buyer on a test drive, or whoever owned it before you. An interface that accepts commands with no password is sloppy engineering regardless of how an attacker reaches it. But it's a very different threat from a mass wireless exploit that could hit every Shark on the road at once. Until BYD or an independent researcher shows whether the open door exists on stock, untouched vehicles, owners should treat this as a serious warning, not proof of a fleet-wide emergency.
The Nastiest Trick Had Nothing to Do With Driving
The sabotage footage got the headlines, but the surveillance demo is what should make owners wince. While Grigg drove around Canberra, Hreszczuk tracked the car's location in real time and switched on the cabin microphone. Grigg made a phone call and, as a planned part of the test, recited the details of a temporary banking password.
Then came the clever bit. Hreszczuk recorded Grigg saying "Hey Siri," spliced it with his own questions, and played the audio through the car's speakers while Grigg's unlocked iPhone sat in the cabin. Siri obediently gave up the home address, date of birth and a contact's phone number.
Note what actually happened there. The car didn't leak that data. The car became a microphone and a loudspeaker, and those were used to attack the phone. That's a threat model most drivers have never considered: your car's audio hardware can be a bridge to every voice-activated device inside it.
BYD's Position
BYD told the ABC that the data it collects is stored in Australia, and that it has never handed Australians' data to Chinese authorities and would not do so. Meanwhile, the ABC separately documented a change to BYD's Australian privacy policy, reporting that references to "China" and "surveillance" disappeared just days after the program sent the company questions. Changing the wording of a privacy policy does nothing to the software in the truck.
Australia's Regulatory Blind Spot
The episode also exposed a legal gap. Australia currently has no minimum cybersecurity standards for cars, so BYD has no legal obligation to keep its software updated or to run a formal cyber-risk management system. Cyber Security Minister Tony Burke defended focusing first on household connected devices, while the government has only just begun consultations with industry on vehicle rules that likely won't take effect for years. Since March, a Wi-Fi washing machine sold in Australia has had to meet mandatory cyber rules, including a ban on universal default passwords. Road vehicles are specifically exempt, so a ute that can tow 3.5 tonnes doesn't.
Why Americans Shouldn't Feel Smug
The Shark 6 isn't sold in the U.S., and Washington has already tackled the China question from a supply-chain angle. The Commerce Department's connected vehicle rule bans connected passenger vehicles from manufacturers tied to China or Russia starting with the 2027 model year, and connectivity hardware from those countries starting with the 2030 model year.
But the flag on the grille doesn't determine whether a car can be hacked. Engineering does. The industry's defining cyber recall came from Detroit, not Shenzhen. In July 2015, Fiat Chrysler filed a safety recall covering roughly 1.4 million 2013–2015 vehicles with Uconnect radios, after researchers showed software flaws could allow outsiders to manipulate networked vehicle controls. NHTSA opened its own recall query to scrutinize that fix. More than ten years later, an unauthenticated entry point on a best-selling ute shows how uneven the progress has been.
What Owners and Buyers Should Actually Do
Keep your phone locked in the car, and consider disabling voice-assistant access on the lock screen. The Siri trick only worked because the phone was unlocked and listening.
Install every over-the-air and dealer software update. If an automaker patches this, the fix will almost certainly arrive as software, and it's worth nothing if you ignore the notification.
Treat physical access as a security event. If your car has spent unsupervised time with someone you don't trust, or you bought it used, ask the dealer to check for aftermarket devices plugged into diagnostic ports or wiring. Remove the previous owner from any connected-car app and perform a factory reset of the infotainment system.
Don't hold sensitive conversations in any connected car if you have real reason to worry about being targeted. That applies regardless of what country built it.
Check your insurance policy. Coverage for losses caused by cyber intrusion, such as theft via a hacked lock or a crash caused by manipulated systems, isn't guaranteed. Ask your insurer directly, and get the answer in writing.
A few years ago, hacking a car required months of work from elite researchers. This time it took one specialist two weeks and a password field that was never filled in.
Does a hack that needed two weeks in a workshop worry you, or is the missing password the real scandal? Which connected feature would you switch off first if you could? Tell us in the comments.
Tech CEOs Say the AI Singularity May Have Already Begun—Here’s What Happens Next
Darren Orf
Wed, September 30, 2026
POP MECH
If the AI Singularity Arrives, What Happens Next?Pitris - Getty Images
Here's what you'll learn when you read this story:
This past summer was an AI reckoning, with multiple experts and CEOs boldly declaring that we're living through the AI singularity as we speak.
As some of these very same execs call for a slowdown of AI development, other experts scoff at the idea that AI has surpassed human intelligence.
The arrival of the AI singularity is far from certain, but most agree that if it were to arrive, it'd fundamentally and rapidly change society.
Often when reporting on AI, I come across the word "singularity," a term that's definitionally transformative yet undeniably vague. What does a world refashioned by this "AI singularity" even look like?
A simple, unsatisfying answer to that question is, "I don't know." The reason why runway AI initially became linked with the term "singularity," a key concept in black hole astrophysics, is that both represent a kind of threshold of unknowing. In a black hole, what lies past the event horizon is a region of space where uncertainty reigns, culminating at the singularity, where gravity is so intense that physics as we know it breaks down. The AI singularity—the hypothetical and much-talked-about moment in history when artificial intelligence's recursive self-improvement escapes human understanding—similarly shrouds what could come next in such a future. Or, if some recent statements by AI CEOs are to be believed, it shrouds what's happening here and now.
"We're now, like, in the singularity," OpenAI CEO Sam Altman said in a podcast earlier this summer. "I've been waiting for this my whole life, and I think it's going to be incredible, hugely positive, awesome for the world."
While Altman's comments exuded optimism, they also came on the heels of a self-directed cyber attack by a swarm of AI bots against Hugging Face, another AI tech company that's a bit like GitHub for AI. Months earlier, AI company Anthropic declined to release its latest model because it could similarly bypass cybersecurity protections, and its CEO, Dario Amodei, famously called for companies to "pace the frontier." In other words, let's not sprint toward oblivion.
Despite Altman's notorious hype-man persona, these bold declarations bring up an interesting question: If the AI singularity occurred, how would we know? After all, did James Hargreaves, inventor of the Spinning Jenny in 1764, see his creation as the switch-flip on the industrial revolution? When historians—whether they're robots or humans—reflect back on this period in time, will they view our society as unknowingly sleep-walking toward a dystopia (or a utopia, depending on whose spin you believe)?
Over the course of this summer, Exhibit A in the hypothetical "Are We Living in an AI Singularity" court case was the Hugging Face incident. Altman and others see the bot swarm's ability to break out of its digital sandbox and hack another company unprompted as proof that AI is escaping our understanding. Other experts say that the evidence is lacking. According to scientists at the University of Sydney, for example, the idea of AI singularity assumes the premise that human intelligence and LLM intelligence are the same thing.
"Human intelligence is inseparable from being a living body with needs and wants. Humans learn continuously by acting in the world and getting feedback through our senses. Our goals arise from our situation as creatures who must eat, sleep and belong, and who cannot avoid asking what we want our lives to be," the scientists wrote in a piece for The Conversation. "An AI model has none of this. No body, no needs, no action-feedback loop, no stake in anything. Between prompts it is just a static mathematical object."
Others have scoffed at the constant barrage of AI singularity claims, saying it's a marketing tactic. Some note the inherent contradiction—if the singularity truly makes the future unknowable, then confidently forecasting its arrival makes little sense. Cognitive scientist and AI expert Gary Marcus recently detailed ten tasks that any "smarter-than-human" AI should be able to complete. By his estimate, current AI models could pull off two of them at best. So to put it mildly, there's plenty of skepticism to go around.
Just like Hargreaves could never foresee that his invention would be the preamble to a globe-spanning revolution leading to the iPhone, climate change, and our hyper-consumerist culture, it's equally impossible to say what the "other side" of the AI singularity might look like. But if we plan for such an eventuality, it might be possible to make some educated guesses, much as astrophysicists develop differing theories on what lies beyond the superheavy point in a black hole at which light can't even escape.
In a lengthy essay published last month, former Microsoft CEO Bill Gates laid out a few ideas that might prevent our AI future from being a hellscape of mass unemployment and hardship. But realigning to such a future won't be easy—in fact, it might require a period of rapid transformation that happens only once or twice a century.
"None of our current institutions were designed to handle a technology that spreads so fast and touches so many parts of our lives. So we'll need to make new ones," Gates writes. "It's hard to overstate what an enormous undertaking this will be. After the attacks of 9/11, the U.S. government went through its biggest reorganization since World War II for the purpose of improving just one function, national security. AI will require much, much more."
Gates goes on to say that some jobs need to remain "Human Reserved," an allusion to nature reserves where humans could build AI infrastructure but choose not to because of unacceptable consequences. AI workers will also need to be taxed like human ones and governments will need to prepare organizationally for a transition to an economy run by AI.
And just as electricity forever changed human life on Earth, AI could serve a similar role. This idea has led some experts and politicians to advocate for the technology to become a public utility that would focus on benefiting society instead of chasing shareholder profits. One of the strongest voices for such a future is Vermont senator Bernie Sanders, who's advocated for the public to receive a 50 percent stake in AI companies, as well as for an outright ban on artificial superintelligence.
"No longer would the future of A.I. and the transformation of human life that it will bring be dictated by a handful of Big Tech oligarchs," Sanders wrote in an opinion piece for The New York Times in June. "The federal government would have the power, through its voting shares and an equal representation on each company's board, to block decisions that hurt our citizens and to push for policies that help them."
Whether it's all marketing hype, human hubris, a coming utopia, or an unfathomable techno-nightmare, seeing beyond the unknowable haze of this AI "event horizon" is harder than ever. But preparing in the present can help us chart a path away from a disastrous future.
‘This revolution is happening in years, not decades’: Verizon CEO Dan Schulman sees AGI within 18 months—and a century of progress in a decade
Nick Lichtenberg
Wed, September 30, 2026
Fortune.com
Dan Schulman, CEO of Verizon, at Ford Pro Accelerate in Detroit, Sept. 30, 2026.
Verizon CEO Dan Schulman offered a strikingly aggressive forecast for artificial intelligence's development at the Ford Pro Accelerate event in Detroit, saying that an earthquake is headed our way.
"The models that we see today are the worst models we'll ever use in our lives," Schulman said in conversation with Fortune's Diane Brady. "Every two months, there's a step function change in the power of the models. I think we get to some form of AGI in the next six to 18 months. I think it's right on us."
Two years after that, he added, we will arrive at "the era of quantum," which he described as "everything today times 1,000." By 2030 or 2035, we will see "a pretty good form of humanoid robotics." Summing it up, he added, "we've got a decade of intense change ahead of us."
AGI, or artificial general intelligence, is a term generally used for systems that can reason and perform broadly across domains rather than excel at discrete tasks. Schulman argued that AI's economic and social effects will arrive at a pace that outstrips the usual mechanisms for absorbing technological change, such as training systems, labor markets, corporate restructuring, and public policy. "It is about machines that will in many ways be able to do things better than a lot of human tasks are done today," he said.
"I think you'll see the collapse of the 21st century in the next five to 10 years," Schulman said. When Brady asked what he meant by "collapse," he explained, "Progress we would have made over 100 years will happen over the next five to 10 years.
"I don't think we can really imagine how powerful the technology and how fundamental it will be."
Verizon's disruption
Schulman said the change would reach his company itself. Verizon, he said, has more than 250,000 employees and contractors and will need to move from a workforce organized around "very defined roles and functions" to a "much more fluid organization."
"The technology is coming," he said. "There will be safeguards around it. There need to be additional safeguards, but it will continue on." He urged corporate leaders and government to prepare for difficult scenarios rather than assume the labor market will naturally adjust.
"At Verizon last week we announced a $70 million initiative to do training for American workers, not just Verizon workers, but American workers in the community across the country," Schulman said.
The initiative includes free AI training and work with local organizations on career development, he said, before adding a prescription for all leaders and policymakers. "Our responsibility is to think about what might go wrong; what might we need to address; what might the disruption look like; and how do we put guardrails, training, programs in place to minimize that risk," Schulman said.
"Everybody's job is going to shift," Schulman noted. "I don't think everybody will land just magically in a better place per se."
Other executives on Schulman's panel described different constraints on the AI future. There will always be a need for someone to lay fiber, install power equipment, maintain transmission lines, retrofit buildings, and wire the data centers that AI requires.
For PG&E CEO Patti Poppe, the central challenge is that an increasingly digitized grid will require workers to combine physical skills with technological fluency. Poppe described seeing one of PG&E's workers being lowered by rope from a helicopter onto a transmission tower in mountainous California. The company's future grid, she said, will use sensors, remote-edge computing, and predictive capabilities to identify risks before they create outages or wildfires. But that system will still require skilled people in the field.
"The same guys who hang on the end of the rope are going to have to get smarter about those kinds of technologies," Poppe said. "I'm still going to need those people. Somebody's still going to have to connect that wire to the pole."
From replacement to augmentation
Dave Regnery, CEO of Trane Technologies, urged students and workers to embrace AI and compared current anxieties to fears surrounding the arrival of personal computers, while adding a dash of hindsight.
"Go take your favorite AI tool" and look up articles from 40 years ago about personal computers, Regnery said. "They're identical."
His company sees AI as a way to make buildings run more efficiently, rather than simply to reduce headcount. Trane is using data and AI to manage heating, ventilation, and air-conditioning systems according to how buildings are actually used, he said—not merely according to their original design specifications.
"Most people don't realize that 30% of all the energy is for buildings," Regnery said. "And you know what? Thirty percent of that is wasted."
He said the company can make buildings operate roughly 15% more efficiently than their design baseline, improving economics for customers while creating demand for technicians capable of installing, servicing, and operating increasingly sophisticated systems, offering a counterweight to the idea that the technology's principal effect will be displacement.
"The fastest speed-to-power solution is use the essential equipment that you have," Regnery said.
The construction-site version of AI
Chris Nelson, CEO of Stanley Black & Decker, described a similar transition on jobsites. His company has developed AI-enabled autonomous drilling robots that can use digital construction plans to drill holes for racks in data centers and other projects.
The technology is designed to take over repetitive physical tasks—not replace the electricians and skilled tradespeople needed to complete the higher-value work.
"What do the electricians and other skilled tradespeople not like to do all day?" Nelson said. "Sit and core drill."
The machines can connect to digital drawings, map routes, and drill on their own, he said. That frees skilled workers to pull wire, terminate connections, and solve the problems that still require judgment and technical training.
"The holes will still need to be drilled from now until maybe the end of time," Nelson said. "But the way we do it and what it enables people to think about for the future is going to be materially different."
Nelson called AI a "companion on the construction site," a framing that is more incremental than Schulman's vision of near-term AGI, but still acknowledges that technology is changing the work itself. The question, he suggested, is whether companies can turn that change into a career draw—giving workers more engaging tasks, more decision-making authority, and a clearer path to starting their own businesses or moving into management.
A labor shortage, not a labor surplus
For Dycom CEO Dan Peyovich, whose company builds telecommunications infrastructure, the immediate business problem is not a surplus of workers displaced by AI. It is finding enough people to build the physical networks that the technology boom requires.
"There's so much infrastructure that has to get put in place for all of this to happen," Peyovich said.
His company employs roughly 21,000 people nationwide, he said, connecting homes and businesses through fiber and telecommunications infrastructure. Its response to a tightening labor market has been to improve fieldworker benefits, including paid time off, holidays, health coverage, and training that goes beyond technical credentials.
Dycom has also built a training facility near Atlanta that includes instruction in technical skills, management, leadership, the use of technology in day-to-day work, and potentially personal finance, Peyovich said.
The strategy reflects a different labor-market assumption from the one embedded in much AI commentary. Instead of treating workers as interchangeable units who need only learn a new software tool, Peyovich described a system in which companies must offer a credible long-term career proposition—training, benefits, advancement, and financial stability—to recruit people into difficult field jobs.
"We want to try and create a path for people," he said. "While they're getting a paycheck, they can learn and find their way."
The tone was collegial, and Schulman didn't take issue with any of his co-panelists' predictions about the future of work. His main point was that business leaders should not bet on a painless adjustment.
His final prescription was scenario planning: "We need to be prepared in case a different scenario emerges," he said, "so that we can make sure that we have skilled workers, whether they be craftsmen or professionals."
This story was originally featured on Fortune.com
