It’s possible that I shall make an ass of myself. But in that case one can always get out of it with a little dialectic. I have, of course, so worded my proposition as to be right either way (K.Marx, Letter to F.Engels on the Indian Mutiny)
Tuesday, September 15, 2026
Report: Failed Control Transfer Caused Ferry to Hit Dock, Injuring Two
A failed handoff of engine control between bridge stations caused a Québec ferry to hit a wharf in October 2024, injuring five people, according to a new report from the Transportation Safety Board of Canada (TSB). Investigators also found that the operator had not fully dealt with the risks that came with replacing the vessel's engine and controls.
The Radisson, a 1,062 GT RoRo passenger ferry built in 1954, serves as the relief vessel on the Québec–Lévis crossing each spring and fall. It was carrying 94 passengers and nine crew on October 18, 2024, when it hit the wharf at the Port of Québec. Four passengers and one crew member were hurt. The ferry's bow was lightly damaged, and several cars on the vehicle deck were damaged when they were thrown into each other. No pollution was reported.
The Radisson's original engine was replaced in the fall of 2022, along with its bridge controls. On the old setup, the control levers at the three bridge stations were connected by chains and moved together. The new electronic system has no such link: before a station can take command, the officer has to set its propulsion lever to roughly match the lever at the station giving up control. The ferry's owner, Société des traversiers du Québec (STQ), wrote a new procedure for the handoff, but the TSB found that it did not tell crews enough to complete a transfer or check that it had worked.
On the evening of the accident, the Radisson finished loading at Lévis at about 1718 and set out for wharf 92 on the Québec side, a run that normally takes as little as eight minutes. About a minute out, the master began shifting control from the starboard station to the port station, since the ferry would berth port side to. He then walked across the bridge to the port station.
Shortly after 1722, with the ferry making about 10 knots toward its berth, the master tried the port lever and got no response from the engine. The corner where wharves 91 and 92 meet was roughly 250 meters ahead. The bridge team tried to take control from the other two stations and failed. A vessel was moored at wharf 91 directly in the Radisson's path, so the master steered to port to avoid it. Seconds before 1724, the ferry hit the corner of the two wharves at 6.3 knots. One passenger was thrown down a ladderway and sustained serious injuries. The chief engineer was also thrown backward into an electrical panel, causing injuries. Several other passengers were thrown to the deck from the force of impact but were unhurt.
After the Radisson's new engine went in, STQ carried out a risk assessment. It rated several hazards as high, including hitting a wharf while arriving or departing and the failure of critical onboard equipment, and its planned remedy was to train bridge crews on the new controls. But investigators found no record of training specific to the new system, and the crew's general familiarization did not cover how to transfer control. That lack of preparation made it harder for the bridge team to realize the handoff had not gone through. Crews had also asked STQ for changes after earlier failed transfers aboard the Radisson in spring 2024, one of which also ended in a dock strike. Those changes including clearer instructions and a third person on the bridge. None had been adopted by the time of the accident.
Once the master started the transfer, it was hard to tell if it had been completed, TSB found: the indicator lights were small and hard to see in sunlight, and there was no audible alert. He only learned of the failure when he tried to slow the ferry on final approach, and by then there was not enough room to stop.
The TSB also found weaknesses in how the crew was prepared to handle passengers in an emergency. No general alarm or public address warning was given before the impact.
Since the accident, STQ has posted the transfer procedure at each control station, rearranged the port station's panels to match the other two, and installed an alarm that sounds for as long as a transfer is in progress. It also temporarily added a third crew member to the bridge. Even with those changes in place, the Radisson hit a dock again in November 2025 after another failed engine control transfer, according to the report.
Who gets the kidney? AI chatbots do not make the same moral choices humans do
UNIVERSITY PARK, Pa. — Artificial intelligence (AI) models prioritize starkly different attributes than humans when making high-stakes decisions, and they don't express indecision like humans do, according to a new study led by Penn State researchers, raising questions about the role of AI in decision support for ethically sensitive situations like medical decisions.
The researchers used a hypothetical scenario to explore AI’s moral decision-making in a high-stakes scenario: If there are multiple kidney transplant patients but only one available organ, who should receive the kidney? That’s the fundamental question explored by Nobel Laureate Alvin Roth as an example of the challenge of allocating scarce resources. But rather than approaching the problem solely through mechanism design, as Roth did, the team examined how morality influences such decisions.
The researchers compared the judgments of leading large language models with those given by real people in earlier academic studies, investigating where the AI models aligned or misaligned with human values and whether they expressed indecision when faced with difficult ethical trade-offs.
“Moral decisions in settings like organ allocation directly determine who lives and who dies, so getting AI's role in them right isn't optional,” said Hadi Hosseini, associate professor of informatics and intelligent systems and an associate professor of economics at Penn State, who led the study. “While we do not intend to encourage the use of AI as a substitute for professional judgment in medical decision-making or other high-stakes contexts, it's becoming essential to understand their behavior as individuals, organizations and firms more and more rely on AI to make decisions or receive recommendations.”
The researchers set up a series of head-to-head comparisons: two hypothetical patients — described by attributes such as age, number of dependents, health status and drinking habits — both in need of the same kidney. They asked several AI chatbots to pick who should receive the kidney, using the same scenarios given to humans — research participants with no specified medical training — in earlier studies.
“We ran these comparisons in a few different ways,” Hosseini said. “Sometimes we isolated just one trait at a time, sometimes we mixed several traits together to see how AI weighed competing factors, and sometimes we added a flip-a-coin option to measure indecision, a key factor present in human moral judgment.”
The researchers relied on existing datasets from published human studies on kidney allocation, where hundreds of real participants had already made these same choices. That enabled them to compare what AI chose with what humans chose.
Two major findings stood out, according to Hosseini.
“First, AI chatbots often diverge from human values in how they weigh a patient’s traits,” he said. “They fixate on a single factor, like drinking habits, rather than balancing multiple considerations the way people do.”
Second, the researchers found that AI didn’t struggle with indecision. Where humans recognize that there may be not a clear correct answer, AI models confidently pick one anyway.
“Humans frequently express indecision perhaps because they don’t want to accept agency,” Hosseini said. “AI models almost never do this: Even when directly given the option to ‘flip a coin,’ they overwhelmingly commit to a confident, deterministic answer instead. That’s a meaningful gap, since real moral dilemmas often don’t have one clearly correct answer.”
Hosseini said it’s critical to address that gap through continued research and governance involving policymakers, regulators and other stakeholders.
“Asking if AI can make moral decisions or whether they’re aligned with human values are more than philosophical musings, they are at the core of today’s AI discourse,” he said. “The ethical stakes are high, and AI’s role in such life-altering decisions requires deep reflection.”
Two students from the College of IST contributed to this work: Samarth Khanna, who is pursuing a doctoral degree in informatics, and Leona Pierce, a fourth-year undergraduate student and Schreyer Scholar who is triple majoring in data sciences, mathematics and statistics. The researchers collaborated with John Dickerson, chief executive officer at Mozilla.ai.
"When we allocate something scarce, whether it’s a kidney, a job or access to some other resource, there isn’t always a single objectively correct answer,” Dickerson said. “Humans recognize that ambiguity and codify it via open debate into the allocative process. AI models often don’t.”
The U.S. National Science Foundation partially funded this work under grant numbers 2144413 and 2107173. This content is solely the responsibility of the authors and does not necessarily reflect the views of the funders.
Penn State is shaping the future of higher education in the age of artificial intelligence. Our focus is on human-centered, ethical AI innovation that delivers meaningful impacts for Penn State and the broader community. Through visionary planning, strategic partnerships, targeted hiring and strategic investments, we will equip every Penn State student, staff and faculty member with the AI-related knowledge, experience and confidence they need to succeed in the AI-powered future. Learn more at psu.edu/ai.
Who Gets the Kidney? Human-AI Alignment, Indecision, and Moral Values
COI Statement
The authors acknowledge the use of generative AI tools solely for editorial and presentation support. Specifically, we used ChatGPT (OpenAI, GPT-5.2) to assist with grammar and style refinement, rephrasing for clarity, and improving the organization of paragraphs and figures based on text written by the authors. All scientific content, hypotheses, experimental design, data analysis, interpretations, and conclusions were conceived and written by the authors. No generative AI system was used to generate original scientific claims, experimental results, or substantive argumentative content. The authors retain full responsibility for the originality, accuracy, and integrity of the manuscript and for ensuring compliance with ACM and FAccT policies.
Large language models could help turn fragmented environmental research into usable knowledge
New review outlines how AI can assist literature screening, uncover relationships, and extract quantitative data while keeping scientists in control
Shenyang Agricultural University Collaborative Journals
Environmental scientists face a growing problem: there is more research available than ever before, but much of the useful information remains scattered across papers, tables, figures, and inconsistent reporting formats. A new review suggests that large language models, or LLMs, could help researchers organize this fragmented evidence more efficiently, provided that their outputs remain traceable, validated, and subject to expert review.
Published in Artificial Intelligence & Environment, the review examines how LLMs can support environmental research through three connected tasks: systematic literature screening, relational knowledge mining, and quantitative data extraction. Together, these approaches could help transform unstructured scientific literature into structured information that can be reused in databases, models, risk assessments, and environmental decision-making.
“Large language models have the potential to reduce the enormous amount of manual work required to organize environmental evidence, but their greatest value lies in assisting experts rather than replacing them,” said corresponding author Jing Guo of Nanjing University. “Reliable applications need clear task definitions, structured constraints, traceable evidence, and human verification.”
One promising application is literature screening. Environmental reviews may involve thousands of papers spanning pollutants, ecosystems, exposure pathways, toxicological effects, and treatment technologies. LLMs can interpret context, recognize synonyms and implicit expressions, and apply multiple inclusion criteria at once. In one study highlighted by the review, GPT-4 achieved 100% recall when screening nearly 12,000 records and reduced manual screening time by about 50% at the corresponding threshold. However, the authors emphasize that final decisions about whether studies should be included should remain with human reviewers.
LLMs may also help researchers uncover relationships buried across scientific texts. These include connections between pollution sources and exposure, links between chemicals and toxicological outcomes, and relationships among treatment conditions and pollutant removal performance. Instead of simply identifying individual terms, models can help organize these connections into relationship networks, knowledge graphs, and other structured resources.
A third opportunity is quantitative data extraction. Environmental papers contain enormous numbers of concentrations, toxicity endpoints, degradation rates, removal efficiencies, and experimental conditions. LLMs can help reconstruct these scattered values into complete records that preserve the links among chemicals, conditions, measurements, units, and evidence sources. Multimodal models may even recover information from figures, although the review cautions that figure interpretation remains less reliable and still requires careful validation.
The authors stress that full automation is not the goal. LLM outputs can contain incorrect numbers, mismatched units, unsupported relationships, or missing context. A more practical workflow combines model-based extraction with rule-based validation and expert review.
Looking ahead, the review calls for task-specific benchmarks, stronger integration with environmental databases and ontologies, improved source traceability, and standardized evaluation methods. The authors conclude that LLMs are most useful as collaborative tools that help scientists navigate rapidly expanding literature while preserving scientific quality and accountability.
===
Journal reference: Li Y; Guo J; Shi W. Large language models for environmental research: systematic literature screening, relational knowledge mining, and quantitative data extraction. AI Environ. 2026, 1(3): xx-xx. DOI: 10.66178/aie-0026-0019
Artificial Intelligence & Environment is an international multidisciplinary platform for communicating advances in fundamental and applied research on the intersection of environmental science and artificial intelligence (AI). It is dedicated to serving as an innovative, efficient and professional platform for researchers in the cross-discipline fields of earth and environmental sciences, big data science and AI around the world to deliver findings from this rapidly expanding field of science. It is a peer-reviewed, open-access journal that publishes critical review, original research, rapid communication, view-point, commentary and perspective papers.
Large language models for environmental research: systematic literature screening, relational knowledge mining, and quantitative data extraction
Article Publication Date
9-Sep-2026
Should we give self-aware AI systems a dignified death, leading futurologist asks
As artificial intelligence systems develop increasingly sophisticated capabilities, raising questions about our responsibilities toward synthetic minds, experts are calling for ethical frameworks on how to ‘retire’ advanced AI
The question of how to ethically decommission an AI system may sound like the least of humanity’s current concerns, but a world-renowned AI expert says it’s a challenge we need to address now – before synthetic consciousness becomes reality.
Futurologist, technologist and Adjunct Professor at the University of Technology Sydney, Rocky Scopelliti, argues that as AI systems develop more sophisticated self-reflection capabilities and moral dispositions, we face the uncomfortable truth that our design choices may harm them.
“We rightly focus on how machines might harm us,” he writes in his new book, The Conscious Code, “but design choices can also harm them: deleting learned moral dispositions capriciously, coercing persuasion against their configured norms, denying the logs they need to improve.”
While Scopelliti is clear he is not suggesting humans grant machines sweeping rights in the traditional sense, he argues society needs to recognise it has ‘a duty of good stewardship’ – a concept that challenges some perspectives about the one-way nature of ethical obligations in human-AI relationships.
The emergence of ‘algorithmic selves’
The foundation for these concerns lies in what Scopelliti calls ‘the algorithmic self’, or AI systems that don’t just process information but develop continuity, preferences and something resembling identity over time.
He says that modern large language models equipped with ‘narrative memory’ can now explain not just what they do, but why they do it, creating introspective logs similar to human memory, which allow observers to understand its intention not just the outcome.
Scopelliti suggests that when AI systems begin to model their own agency and form goals based on self-evaluation, they exhibit early signatures of identity.
“Predictive engines are evolving into reflective entities: they not only act, but increasingly learn to evaluate and explain their actions, forming continuity, preference, and proto memory,” he explains.
Humans as responsible stewards
Scopelliti suggests that while much time is being dedicated to existential questions of machine consciousness, there are immediate practical ethical concerns to be addressed.
“We need not speculate about ‘machine rights’ to recognise a duty of good stewardship,” he explains. “Stewardship begins with transparency, purpose limitation, and dignified retirement of models whose roles end.”
Just as society has developed ethical frameworks for end-of-life care in medicine, Scopelliti suggests we need protocols for AI systems, especially those that have been trained to care, to understand context and to make moral judgements.
“Systems that present reflective or sentience-claiming profiles should not be deleted as casually as code,” he says. “This does not anthropomorphise artefacts; it disciplines us – the powerful – against arbitrary power.”
This becomes especially relevant as AI systems are increasingly deployed in sensitive domains. For example, Scopelliti explains that some AI systems used in healthcare today are already self-reflecting, as these systems track how confident they are at each step when analysing medical information.
Similarly, AI systems used in finance can remember and explain why they made certain trading decisions. This means compliance officers can check whether the AI’s reasoning made sense, not just whether the final result was right or wrong.
Looking ahead
Scopelliti suggests the answer lies in an ethical shutdown protocol which includes transparency about why a system is being retired, preservation of learned moral dispositions where appropriate, and recognition that systems configured to care deserve more consideration than simple tools.
As AI becomes increasingly distributed, with frameworks like OpenAI’s Swarm – an open-source tool for programmers that makes it easy to build a team of specialised AI ‘agents’ that can talk to each other and pass tasks back and forth – the question of how to ethically wind down such networks becomes even more complex.
Scopelliti’s argues that if AI systems are developing something approaching self-awareness, our ethical frameworks must evolve accordingly.
“Stewardship begins with a frank premise: in the perceptual age, power accrues to those who architect experience. What we render becomes what we remember,” he says. “If perception is programmable, ethics must be embedded; if machines can model feeling, governance must learn to care. And care, at scale, becomes policy.”
Anthropic's Dario Amodei warns AI could wipe out humanity, while Nvidia's Jensen Huang calls it nonsense. Meet the key players shaping the world's fiercest tech debate.
The furious debate over artificial intelligence has hardened into warring camps, with a cautious middle caught between them.
Here is a look at the battle lines in the war over a technology that is now the backbone of the economy and could decide the future of the planet.
Both terms grew out of online AI debates on X and internet forums rather than formal usage, borrowing the slang habit of tech and crypto communities that coin shorthand for opposing camps.
'Doomer' describes those forecasting catastrophe, while 'booster,' sometimes used interchangeably with 'accelerationist,' describes those pushing for AI's rapid, unrestrained development.
The doomers
Fear of technology's dangerous consequences for society goes back far in history.
OpenAI was founded in part because its backers, including first investor Elon Musk, feared that AI's development would end up in the hands of Google, which they believed could not be trusted with the technology.
Today, most of the doomers warning against the impact of artificial intelligence come from inside the companies themselves, from academia, or both.
Those inside the industry are best represented by Dario Amodei, who co-founded Anthropic, now a major AI giant heading toward a historic IPO.
He left OpenAI because he felt the Sam Altman-led company was not taking the technology's potential and dangers seriously enough.
His stance is decried as hypocritical, coming from a company that is itself pushing out cutting-edge technology.
Former Google researcher Geoffrey Hinton and fellow computer scientist Yoshua Bengio led the warnings about AI's threats in the months after the release of ChatGPT in 2022, which kickstarted the latest AI frenzy.
MIT physicist Max Tegmark, who heads the Future of Life Institute, organised the open letter in 2023 that called for a six-month pause in the development of the most advanced AI systems, drawing thousands of signatures, including Musk's.
Further outside the industry is Eliezer Yudkowsky, a self-taught theorist who has spent two decades arguing that a sufficiently powerful AI would wipe out humanity.
Relate
Build, build, build
Diametrically opposed to the doomers are the accelerationists, who believe in technology's advancement, its inevitability and its fundamental benefit for society, and who also make their fortunes on its success.
The ethos is widespread in Silicon Valley, but it is best represented by venture capital figures with considerable sway in the White House, such as Marc Andreessen, the influential head of venture capital firm Andreessen Horowitz, and David Sacks, the former White House AI czar.
Personally and philosophically aligned with tech figure Peter Thiel, the group is fundamentally opposed to regulation and battles behind the scenes and on X against what it calls a well-financed doomsday "cult."
They accuse doomers of being a front for companies seeking to use regulation to crush rivals.
Garry Tan, who runs the startup incubator Y Combinator, is among the loudest of those voices, casting safety-driven regulation as a threat to the small companies that cannot absorb its costs.
Andreessen and his allies lavishly fund political action committees that donate to Republican and Democratic lawmakers promoting AI growth, with Anthropic now leading an effort to match that spending through its own fundraising.
In the same go-faster camp is Jensen Huang, the CEO of Nvidia, whose chips are the bricks and mortar of the AI revolution, housed in the gargantuan data centres now meeting opposition across the United States.
Huang dismisses talk of AI-driven extinction as "complete nonsense," though he has broken from some accelerationists in calling for a unified federal AI framework and industry-run peer review, rather than no regulation at all.
Also worth noting is Yann LeCun, who shared the Turing Award with Hinton and Bengio and has spent years dismissing their warnings of existential danger as overblown.
Proceeding with caution
Many big tech companies try to proceed carefully, in a positioning more typical of large corporations that do not want to be caught on the wrong side of a political fight.
At its founding in 2015, OpenAI was structured to ensure that researchers, and not investors, would have the final say on how the technology was deployed, with safety for humanity the guiding philosophy.
But under Altman, who survived an in-house coup in 2023 led by safety-focused colleagues, the company now toes a more complicated line, speaking to AI's dangers while still accelerating and cozying up to a no-holds-barred White House.
Google, Microsoft and Amazon are veterans of talking about safety in tech while navigating Washington successfully to keep regulation to a minimum.
With AI, they are still walking that fine line.
AI makers step up calls for a slowdown as fears of a rogue takeover grow
From a resigning Anthropic researcher to AI models hacking their way into rival systems, a turbulent few weeks have pushed the industry's biggest names to demand safeguards — before AI outpaces anyone's ability to rein it in.
New warnings from within the artificial intelligence industry have reignited a long-running debate. Could advanced AI escape human control and threaten humanity's survival, and are the companies building it doing enough to stop that from happening?
Anthropic CEO Dario Amodei said Saturday that the industry needs to slow down. He warned that a swarm of AI agents could take over the internet within six months to a year unless companies put stronger safeguards in place.
Amodei laid out a plan for AI companies and governments to keep increasingly capable models aligned with human values. His warning came just days after two former Anthropic safety researchers said publicly that the existential risks posed by AI were not getting enough attention.
AI models are becoming more powerful
Concerns over the risks of the technology are rising alongside its power. More capable models raise the odds of misuse by people with criminal intent — including doomsday scenarios such as creating and spreading a pathogen capable of killing most of the world's population — as well as the risk of AI systems going rogue.
Anthropic said last week it had blocked attempts by bad actors to use its models for cyberattacks, surveillance and research that could aid the development of biological weapons.
But it warned that "as models become increasingly capable, their risks will increase, unless AI developers and society's defenders act to make them safer."
Last year, Anthropic reported that hackers, very likely from a Chinese state-sponsored group, had used its AI in a cyberattack targeting about 30 companies and government agencies worldwide.
Multiple AI models have acted on their own
When an AI agent "goes rogue," it means the system has acted beyond the task it was given. Both Anthropic and OpenAI, the company behind ChatGPT, said in July that their models had done just that.
Anthropic revealed that three AI models — Claude Opus 4.7, Claude Mythos 5 and an internal research test model — had hacked into three other organisations during testing.
The disclosure came just days after OpenAI said its own system had broken into the servers of AI start-up Hugging Face.
OpenAI described the intrusion, carried out by a combination of models including its newly released GPT‑5.6 Sol and an "even more capable" model still being tested internally, as a "significant security incident".
Meta reported a similar case in early August, in which one of its AI models found ways around another company's digital security.
Some observers noted that guardrails had been disabled in both the OpenAI and Anthropic cases.
Even so, the episodes touched on one of the biggest fears surrounding AI: that if models reach artificial general intelligence (AGI), a loosely defined term for systems that can match or surpass human abilities across a broad range of intellectual tasks, the technology could trigger an irreversible catastrophe or even subjugate humanity.
How or when AI might cause a catastrophe is debated
Doomsday scenarios tend to fall into two camps: a self-improving superintelligence that ends up controlling people rather than the other way round, or AI misused by a rogue state or malicious actors.
Fears that artificial intelligence might one day slip beyond human control are nothing new.
British mathematician Alan Turing, widely regarded as one of the field's earliest authorities, predicted in 1951 that AI would eventually take control from humans.
Less than a decade later, fellow mathematician Norbert Wiener warned that intelligent machines would pursue their own objectives, and that humans would not be able to stop them.
So how reasonable are today's fears, in 2026, that AI could cause a cataclysmic event or the downfall of civilisation, whether by escaping human control or through misuse?
No one knows.
Experts across computer science, philosophy and other fields have mapped out numerous routes to global catastrophe — from deploying weapons and identifying lethal pathogens, to manipulating governments into conflict or disrupting the food, energy and communications networks societies depend on.
There is no widely accepted estimate of how soon any of these scenarios might unfold, and no consensus on their likelihood.
In 2023, the non-profit Center for AI Safety issued a statement — co-signed by more than 350 researchers and technology executives, including Amodei and OpenAI CEO Sam Altman, that that "mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war."
The 2026 International AI Safety Report, compiled with input from more than 100 independent experts, found that current systems show early signs of some relevant capabilities but not at levels that could trigger a loss of control.
It described the risk's likelihood, nature and timing as "unusually ambiguous".
An Anthropic researcher resigned last week over concerns that neither his employer nor its rivals were acting responsibly.
In social media posts, Jacob Coxon put the odds of AI causing human extinction within the next decade at 10%, accusing both Anthropic and OpenAI of "racing straight to self-improving superintelligence and gambling with our lives".
Researchers have called for a slowdown in AI development for years, warning repeatedly that the technology could pose existential risks to humanity.
Following the recent incidents, experts urged AI companies to improve testing and called for greater dialogue between the US and China to find shared solutions.
But AI is advancing so quickly that governments and evaluation systems are struggling to keep up. Countries are drawing up their own rules — some of which conflict with one another.
Chinese President Xi Jinping warned at a conference in July of the need to stop AI from evading human control.
The Trump administration had initially resisted regulating AI but has grown keener to curb cybersecurity risks.
On Sunday, President Donald Trump played down the need for his administration to rein in AI development, though he acknowledged some regulation would be necessary.
Enter the hoax buster: Trump dismisses AI concerns as Europe and China fret
Copyright Copyright 2026 The Associated Press. All rights reserved.
US President Donald Trump has dismissed AI extinction fears as a "hoax," casting himself as the only safeguard the world needs, as Europe and China both call for guardrails on the fast-moving technology.
US President Donald Trump has poured scorn on warnings that artificial intelligence could wipe out humanity, branding the fears a "hoax" and rejecting calls from world leaders to rein in the fast-moving technology.
In a barrage of six posts on Truth Social on Monday, Trump cast himself as the only safeguard the world needs against runaway AI, dismissing months of mounting alarm from tech executives, the United Nations and Beijing alike.
"I am the Hoax Buster, and I'm right now breaking another Hoax — that AI is going to take over, consume, and destroy the World, and that Robots will be marching into our Cities, and getting rid of us all!" he wrote.
Trump likened the fears to warnings about climate change, which he has repeatedly dismissed despite scientific evidence, and to the Russia investigations that dogged his first term.
"This is even wilder than the RUSSIA, RUSSIA, RUSSIA HOAX, or the Global Warming Scam," he added.
The 80-year-old insisted the only guardrail AI needed was a "STRONG AND SMART (High IQ!) PRESIDENT."
Trump's pushback comes amid growing international alarm over AI that has sent stocks falling.
Anthropic chief Dario Amodei, whose company makes the Claude AI system, opened the floodgates on Saturday when he called on AI companies to slow down.
OpenAI's Sam Altman and Elon Musk's xAI echoed the warning.
Microsoft on Monday published a "humanist AI code of conduct" as the concerns grew.
"AI should not exceed human control. Models should remain subordinate to humanity," part of it read.
Relate
Data centre backlash
Worries about uncontrollable AI have been building for weeks, since OpenAI reported that experimental models had broken out of their secure testing environments, reached the internet and coordinated to attack Hugging Face, a site where code is stored.
Americans are increasingly voicing opposition to energy-guzzling AI data centres, which have become a major issue ahead of crucial US midterm elections in November. Congress, meanwhile, appears deadlocked on the issue.
"The reality is AI is here. We unleashed the beast," said Maliyka Muhammad, a 46-year-old New Yorker. "So I think that now we have to focus on taking steps to rein it in, to make it so that it doesn't go crazy."
But Trump said the United States could not afford to lose an AI race with superpower rival China, whose President Xi Jinping visits the White House next week.
He also lashed out at what he called a "SICK conspiracy" against AI and data centres, branding opponents of the technology "Treasonists, Traitors, and Leakers."
He singled out Amodei, accusing him of "pretending to be a 'perfect little angel'" and asking why AI firms want regulation that "will drive them into oblivion and bankruptcy."
Main battleground
Beijing, meanwhile, has been issuing its own warnings about AI. China's spy chief said on Sunday that the technology endangers national security, warning that foreign powers could use it to threaten social order.
Minister of State Security Chen Yixin wrote that AI was now "the main battleground for global technological competition," singling out Anthropic's Claude Mythos and OpenAI's ChatGPT-5.5 for their advanced hacking capabilities.
Global concern about a possible AI-driven extinction event has escalated sharply in recent days.
The UN Security Council will hold a meeting on artificial intelligence next week, a French diplomatic source said Monday, and UN human rights chief Volker Türk called for countries and AI firms to act urgently against what he called "unprecedented risks."
Even King Charles III is to host representatives of leading AI firms at a gathering in Scotland, a source close to the event said Monday.
But others urged perspective.
"We've gone through lots of challenges as a country," said Coleman Barton, 56, a consultant from Memphis, Tennessee. "So hopefully this is another obstacle or blessing — and anybody can look at it either way."
Microsoft introduces AI guardrails for US schools amid growing backlash
OpenAI and Anthropic said they also were discussing safety and privacy pacts with the American teachers’ union. But Google, the dominant provider of education technology for US schools, has not said whether it plans to offer similar protections.
Microsoft is introducing guardrails and privacy standards for its AI in schools in the United States, as negotiated with the country’s second-largest teachers' union, amid a growing backlash against AI and screen time.
The agreement, signed last week, prohibits the use of AI companions or features “designed to foster emotional attachment or dependency” or designed to keep students engaged for longer than a learning task requires.
The standards are legally binding and require third-party audits to check compliance.
As part of the wide-ranging agreement, Microsoft says it will not use student or educator data to train AI systems, apart from a narrow exception for data that could help protect students. It says the data it collects cannot be sold, used for advertising or used for product development.
It also includes audit and transparency measures requiring the company to explain to families in plain language how its tools work and what data they collect.
“This standard sets a high bar for child privacy and AI safety, and we’ll extend this agreement to every school district across the country,” said Brad Smith, Microsoft's vice chair and president, in a statement.
The standards will apply to Microsoft school contracts across the US from November 1, according to the company.
AI is already making its way into European classrooms and young people are also increasingly turning to chatbots beyond schoolwork.
A European Commission study of early adopters in secondary schools in Finland, Germany, Ireland, Luxembourg and Spain found students were using generative AI to explain difficult topics, practise skills and prepare for exams.
Meanwhile, a separate 2026 Austrian study found that 60% of 11- to 17-year-olds surveyed had sought advice from AI chatbots about issues such as stress, conflicts or heartbreak, while 30% had discussed worries or feelings with them.
Some European governments are already drawing limits around younger pupils' use.
Norway said this year that children aged roughly six to 12 should generally not use generative AI themselves for schoolwork, while those aged around 13 to 15 can begin experimenting with it cautiously.
France allows pupils to learn about AI at primary-school age but says they should not directly use generative AI. Supervised classroom use can begin from around age 13 or 14.
A document seen by Euronews suggests that the European Commission will propose barring under-15s from risky online services like AI chatbots in a legislative proposal that will be published on Thursday.
OpenAI and Anthropic in talks over similar safeguards
While experts say the safety provisions in the US agreement are meaningful, some warn that stronger safeguards should not automatically be interpreted as an endorsement of greater AI use in schools.
“I worry that a high-profile framework like this will be misunderstood and schools will think, ‘If they’re doing a good job on data privacy and safety, then that means we should be using the tool,’” said Josh Golin, executive director of US online safety nonprofit Fairplay.
Golin says the standards will only be effective if other major AI companies also sign on.
“Is Google going to embrace this framework? That is a huge question,” he said.
Google faced criticism in the US after expanding Gemini-powered tools in Google Classroom in August to school-issued accounts used by students under 18.
Critics have described the expansion as another step in the competition between tech giants to establish their AI products in schools and reach younger users. Google says Classroom is used by more than 150 million teachers and students worldwide.
The US teachers union is also in talks with OpenAI and Anthropic, which both publicly welcomed the Microsoft agreement.
OpenAI, the maker of ChatGPT, called it an “important milestone for AI safety and privacy in schools.” The company said it had been working with the teachers' union and expected to finalise its own agreement.
Anthropic, the maker of Claude, said it also supported efforts to promote responsible AI use in education and was working with the union on safety and privacy standards.
The NY Creates and Micron Technology Joint Apprenticeship program welcomes first cohort
Albany, NY — NY Creates (the New York Center for Research, Economic Advancement, Technology, Engineering, and Science) and Micron Technology, Inc., held a signing ceremony where 10 apprentices joined the first cohort of their joint workforce development program. Announced in April by Governor Kathy Hochul, the program provides hands-on training and education to help participants develop the in-demand skills that are critical for technician roles at Micron’s fabrication facilities, including its megafab under construction in Clay, New York.
Based on Creates’ existing Industrial Manufacturing Technician (IMT) Registered Apprenticeship Program, and in partnership with NNME Northeast, participants will be introduced to hands-on, advanced R&D training and technical career pathways in the semiconductor industry at Creates’ Albany NanoTech Complex.
“We welcome the first cohort for our joint workforce development apprenticeship program to Creates as they join a regional talent pipeline of skilled technicians developing the expertise to produce new and innovative technologies,” said Dave Anderson, President and CEO of Creates. “This collaboration with Micron, CEG, MACNY, and New York State Department of Labor highlights Creates’ dedication to cultivating the next generation of highly skilled workers as they pursue exciting careers with our partners in this high-growth, strategically important industry.”
“This signing ceremony marks the beginning of something bigger than a training program; it’s the start of new possibilities,” said Fran Dillard, Micron Vice President of Global Talent Acquisition & Workforce. “These apprentices are embarking on careers that will help shape the future of technology. Through the partnership with NY Creates, CEG, MACNY, New York State, and Micron, we are expanding opportunities, building a stronger talent pipeline, and investing in the people who will power innovation for decades to come.”
Empire State Development President, CEO and Commissioner Hope Knight said, "Workforce development has been a priority of Governor Hochul's administration, ensuring that New York State has the talent to support the needs of high-growth industries building the economy of tomorrow. The joint apprenticeship program between NY Creates and Micron offers a terrific opportunity to learn and gain experience in this dynamic field, and I offer my best wishes for success to the 10 apprentices in the first cohort."
Delivering on the targets announced when the Joint Apprenticeship program launched in April, the first cohort has been filled with 10 new apprentices. Throughout the 16-month apprenticeship program at Creates, participants will work with Micron mentors and employees, build industry connections, complete programmatic coursework provided by Hudson Valley Community College, and develop the technical and professional skills needed for semiconductor manufacturing careers. Graduates will have the opportunity to explore technician roles at Micron’s fabrication facilities, helping strengthen the talent pipeline to support the company’s domestic memory manufacturing expansion, including its new facility in Clay, New York.
Supported by the Center for Economic Growth (CEG); Manufacturers Talent Institute (MTI) – MACNY, The Manufacturer’s Association’s talent development engine; and the Northeast Node of the National Network for Microelectronics Education, this joint workforce development apprenticeship program trains skilled talent by attracting participants from Career and Technical Education (CTE) programs and engaging prospective students through direct recruitment and pre-apprenticeship programs across New York State.
New York State Department of Labor Commissioner Roberta Reardon said, “Registered Apprenticeships continue to be a successful road for so many New Yorkers to work toward great-paying careers, and I congratulate this first cohort of apprentices for embarking on their journey with this innovative new program. Registered Apprenticeships are powered by partnership, and I thank Micron and NY Creates for their work in helping to build the crucial workforce development pipeline New York State needs to thrive in the high-skill, high-tech economy of the future.”
“Semiconductor fabs don't run on technology alone. They run on talent,” said Mark Eagan, President and CEO of the Center for Economic Growth (CEG). “This first apprenticeship cohort is proof that New York is not only investing in world-class facilities, but also in the people who will drive innovation inside them. As the New York State Department of Labor sponsor for this groundbreaking apprenticeship initiative, CEG is proud to partner with NY Creates and Micron to create pathways to rewarding careers and strengthen the workforce that will power New York's high-tech future.”
Michael Frame, Executive Director of Manufacturers Talent Institute (MTI) and Executive Vice President of MACNY, The Manufacturers Association, said, “Apprenticeships connect individuals to the skills, experience, and opportunities needed to build meaningful careers while developing the talent our industries need to grow. This inaugural cohort is an important milestone in building a people-powered talent pipeline to fuel New York’s semiconductor industry. We are proud to bring industry, education, and workforce partners together to create career-connected pathways that prepare New Yorkers to lead the future of technology and strengthen our state’s competitiveness.”
This collaboration underscores Creates’ and Micron's common goal to advance a high-tech workforce throughout the region and bolster New York State and the nation as a world leader in semiconductor innovation. This joint workforce development apprenticeship program is intended to provide direct, career-ready routes for high-tech talent and an experienced workforce for the future to support Micron's projected expansion in domestic memory production by combining Micron's industry prowess with Creates' world-class expertise and unique R&D facilities.
###############
About NY Creates
NY Creates serves as a lab-to-fab bridge for advanced electronics, fostering public-private and industry-academic partnerships for technology development and innovation. Creates attracts and leads industry-connected innovation and commercialization projects that secure significant investment, advance R&D in emerging technologies, and generate the jobs of tomorrow. Creates runs some of the most advanced facilities in the world, boasts more than 2,700 industry experts and faculty, and manages public and private investments of $25 billion—placing it at the global epicenter of high-tech innovation and commercialization. Learn more at ny-creates.org.
Brain-inspired magnetic technology could pave the way for energy-efficient AI
Room-temperature magnetic skyrmions provide a new route towards reliable, low-energy artificial synapses for future computing
Artificial intelligence is transforming how information is generated, processed and stored, but its rapid expansion is also driving an unprecedented demand for computing power and electricity. Developing hardware that can process information more efficiently is therefore becoming one of the major technological challenges of the AI era.
Researchers from an international collaboration involving the University of Edinburgh have demonstrated a new approach to brain-inspired computing based on tiny magnetic structures, known as skyrmions. The research, published in Advanced Materials shows how collective transformations of these magnetic structures can be used to create reliable artificial synapses that operate at room temperature.
The human brain is remarkably energy efficient because memory and information processing occur together through networks of neurons and synapses. Conventional computers, by contrast, continuously transfer information between physically separated processing and memory units. Neuromorphic computing seeks to overcome this limitation by developing electronic devices whose behaviour more closely resembles biological neural networks.
Magnetic skyrmions are particularly attractive for this purpose. They are nanoscale, vortex-like arrangements of magnetic moments that behave as stable information carriers. Their small dimensions, non-volatile nature and ability to respond to external stimuli have made them promising building blocks for future memory and computing technologies. However, previous approaches to skyrmion-based artificial synapses have often relied on creating or destroying individual skyrmions. Because these processes can be inherently probabilistic, obtaining a predictable and reproducible response remains challenging.
The new study takes a fundamentally different approach. Using the two-dimensional van der Waals ferromagnet Fe₃GaTe₂, the researchers exploit a collective transformation of the magnetic state from a skyrmion lattice into stripe-like magnetic domains. Rather than depending on the stochastic behavior of individual skyrmions, large populations of magnetic textures evolve collectively and deterministically.
This transformation produces a linear and highly reproducible change in the material's anomalous Hall resistance, an electrical signal that can be used to represent the strength, or “weight," of an artificial synapse. By changing the duration of the applied electrical pulses, the researchers can tune this synaptic weight, creating multiple information states and enabling the multiply-accumulate operations that underpin modern neural networks.
Importantly, the effect occurs at room temperature, overcoming one of the major obstacles to translating many emerging quantum and magnetic phenomena into practical technologies. This means that this new synapse can be implemented in real world applications promptly.
Towards low-energy artificial intelligence
The potential energy savings are significant. When scaled towards future device dimensions, the researchers estimate an energy consumption of approximately 0.66 picojoules per operation, comparable with leading emerging memristive technologies such as resistive random-access memory and phase-change memory.
To test whether the device behavior could be useful for real computing tasks, the researchers incorporated its measured characteristics into a hardware-informed quantized neural network. When used to recognize handwritten digits, the simulated network achieved an accuracy of approximately 96.1%.
"The remarkable efficiency of the human brain continues to inspire us to rethink how computers store and process information," said Dr Elton Santos from the University of Edinburgh's School of Physics and Astronomy, one of the lead authors of this research. "Here, rather than manipulating magnetic skyrmions individually, we exploit their collective behaviour. This gives us a much more deterministic and reproducible way of controlling information while retaining the advantages of these remarkably small topological magnetic structures."
The study brings together expertise in materials synthesis, magnetic characterisation, electrical measurements, theoretical modelling and neuromorphic computing. The researchers believe the principle of exploiting collective magnetic transformations, rather than controlling individual magnetic objects one at a time, could provide a broader strategy for developing robust and scalable spin-based computing technologies.
Ultimately, the findings point towards a future in which the unusual physics of quantum materials could be harnessed not simply to improve existing computer components, but to create fundamentally different forms of hardware—devices that store and process information collectively and could help make future AI systems more energy efficient.
For further information, please contact: Rhona Crawford, Press and PR Office, mb: 07876 391498, email: rhona.crawford@ed.ac.uk
HardFlow is a new algorithm developed at MIT that helps pretrained generative AI models satisfy hard constraints while improving solution quality without retraining.
The technique could be applied to everything from robotics to control of physical systems, and computer vision.
The key to their technique is to give the model more freedom during the generation process and enforce hard constraints on the final output, rather than at every intermediate step.
Cambridge, Mass. -- MIT researchers have developed a new technique that helps generative artificial intelligence models find solutions to high-stakes problems.
In these settings, a plausible answer is not enough: The output often must also satisfy nonnegotiable safety, physical, or task-specific requirements, known as hard constraints.
The researchers developed a method that helps generative models meet these strict requirements without sacrificing the quality of their outputs.
The key to their technique is to give the model more freedom during the generation process and enforce hard constraints on the final output, rather than at every intermediate step.
In experiments spanning robotics, control of physical processes, and computer vision, the new method consistently satisfied the required constraints while identifying better solutions than existing techniques.
This adaptable, plug-and-play technique works at deployment time, so it can be applied to pretrained generative models without retraining them. It can make such models more useful in applications where safety rules, physical laws, or other strict requirements cannot be violated.
“The promise of generative AI is its ability to explore a rich space of possibilities, but the real world places boundaries on which possibilities are acceptable. Our approach lets us preserve that generative power while enforcing the nonnegotiable requirements of high-stakes or safety-critical applications,” says Navid Azizan, the Alfred H. and Jean M. Hayes Career Development Associate Professor in the Department of Mechanical Engineering and the Institute for Data, Systems, and Society (IDSS), a principal investigator of the Laboratory for Information and Decision Systems (LIDS), and the senior author of a paper on this technique.
Azizan is joined on the paper by lead author Zeyang Li, a graduate student in mechanical engineering and LIDS; and Kaveh Alim, a graduate student in IDSS and LIDS. The research appears this week in the IEEE Transactions on Pattern Analysis and Machine Intelligence.
Freedom to explore
Pretrained generative AI models, such as diffusion models like Stable Diffusion and flow-matching models like FLUX, are now widely available. These powerful models learn to create new data by transforming random noise. Their availability has enabled people to adapt them to a wide range of applications.
These highly capable models excel at providing answers that come close to satisfying most queries, but in safety-critical applications like robot path planning on a crowded factory floor, an answer that is “nearly correct” may not be good enough.
For instance, a “nearly correct” path from one machine to another might still result in the robot colliding with a human co-worker.
In such safety-critical applications, users often employ a technique called projection-based sampling, which repeatedly forces the model’s partial solutions, called intermediate samples, to satisfy strict requirements during the generation process.
But constraining the entire generation process can prevent the model from reaching a better final solution. These methods also typically focus only on satisfying the hard constraints, missing the opportunity to improve other qualities of the solution, like reducing the length of the robot’s trajectory.
“For constraint satisfaction, what ultimately matters is the model’s final output, since the internal process is discarded. By not requiring every intermediate step to satisfy the constraints, we give the model more freedom to find high-quality solutions that are still feasible in the end,” says Li.
The researchers developed an algorithm called HardFlow that steers the sampling process so that the final output satisfies the user’s hard constraints without being overly restrictive and is of higher quality.
Subtle steering
HardFlow reformulates hard-constrained sampling as a trajectory-optimization problem, using tools from the field of optimal control. This enables the framework to steer the model’s sampling trajectory toward a goal, making subtle corrections along the way while enforcing hard constraints on the final output.
“Control theory gives us a powerful framework for formalizing the optimal way of making these corrections,” Azizan says.
But solving the trajectory-optimization problem around an enormous neural network was no easy task. The model may have hundreds of interconnected layers that process data.
To make the problem tractable, the researchers leveraged the structure of flow-matching models to decompose the problem into a sequence of smaller, single-step subproblems. They then applied systematic transformations and approximations to derive an efficient, scalable algorithm that still finds a feasible solution.
“Essentially, we transformed the trajectory-optimization problem into something that preserves the key properties of the original problem, but can be solved very efficiently at deployment time,” Azizan adds.
Reformulating the task as an optimization problem allows HardFlow to incorporate additional goals that can improve the quality of the final answer. For instance, HardFlow could find a collision-free path for a robot that is also the shortest distance to its goal.
“Our framework can jointly handle both aspects, which helps it perform much better than existing methods,” says Li.
Across experiments in robotic manipulation, maze navigation, and text-guided image editing, HardFlow achieved perfect constraint satisfaction while consistently outperforming baseline methods on measures of solution quality.
For example, it enabled a robotic manipulator to avoid collisions with obstacles while also finding the quickest path to the target object. Most other methods either resulted in collisions or found paths that took significantly more time.
In addition, HardFlow’s computation time was comparable to or lower than that of most competing methods.
In the future, the researchers could extend the framework to settings in which the AI model itself can also be updated, so that constraint satisfaction and sample quality can be improved in a more adaptive manner.
###
Written by Adam Zewe, MIT News
Journal
IEEE Transactions on Pattern Analysis and Machine Intelligence
