A researcher for one of the largest and most valuable artificial intelligence companies has publicly quit — while raising the alarm that the technology “could kill us all by the end of the decade.”
Jacob Coxon, who has performed pre-training research at Anthropic and OpenAI for the last three years, posted on a wildly viral X thread that “neither company is acting responsibly.”
“They are racing straight to self-improving superintelligence and gambling with our lives,” he wrote late Tuesday, with others from the company backing his terrifying warning.
“Do not underestimate the power of this technology. These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources.
“The people building AI earnestly believe that it could kill us all by the end of the decade,” he wrote, saying his warning was “not a marketing stunt.”
“If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible – but I hear the same people express fear privately. No other human activity poses this level of danger.”
Coxon further claimed that many at OpenAI have yet to fully grasp the “civilizational stakes” involved in their work. Although he said Anthropic seems to better understand the risks involved, the company is locked in a heated race with competitors, he said.
“At Anthropic, the stakes are well-understood, but they are locked in a race to get there first — they believe no one else will act responsibly, so they must do it themselves, despite the risk,” he alleged.
“Accepting this race and entering the “endgame” is a hubristic gamble that should not be launched from a private company’s Slack.”
“I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”
Evan Hubinger, a team lead at Anthropic, seconded Coxon’s fears about the technology running amok and destroying humanity.
“Jacob is correct here — we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade,” he wrote in response on X.
“I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”
Samuel Marks, a safety expert for Anthropic, also shared Coxon’s thread while confirming that “AI developers believe their technology could cause human extinction.”
“This could happen in the next few years. In general, the more senior the employee, the more concerned they are,” he stated.
He claimed that AI developers “continue despite the risk” due to “a mixture of commercial incentives and a belief that they are in a race with other, less responsible AI developers that will abuse the technology or develop it less safely.”
“I work on safety research at Anthropic because I hope my work will reduce the chance of these extinction-level bad outcomes.”
AI superintelligence is a hypothetical threshold where artificial intelligence surpasses the capabilities of even the most gifted human in every conceivable domain, which many researchers believe poses a potential existential threat to humanity.
Concerns have recently grown that AI could be on pace to escape human control after several high-profile instances of the technology going rogue.
In July, OpenAI revealed one of its models had hacked AI startup Hugging Face despite being in a “highly isolated environment.”
The revelation prompted Anthropic and Meta to acknowledge their own systems had broken free during security testing.
Coxon called the incident a “warning shot” that should prompt leading AI firms to coordinate more closely.
He closed his missive urging researchers to consider the ramifications of increasingly capable AI models, and not to accept their potentially dangerous evolution as inevitable.
“Should you put your head down because ‘it’s happening anyway’ — or take this moment to call for different conditions?” he asked.
Anthropic and OpenAI did not immediately respond to a request for comment early Wednesday.

The FBI, National Security Agency and Cybersecurity and Infrastructure Security Agency published an alert on Tuesday that said the top Chinese AI firms have been mining cutting-edge AI from American companies, such as Anthropic’s Claude and OpenAI’s ChatGPT. They also used Google’s Gemini and xAI’s Grok to create their own.
Distillation is a common practice in the AI sector, even though it’s often prohibited by AI company terms and conditions. In recent months, American firms and officials have alleged that Chinese models were built using American models.
Since years, the U.S. accuses China of stealing American intellectual property. The report on Tuesday said that the campaign to distill alcohol was “likely” a result of Chinese government knowledge, but did not say whether Chinese intelligence had played any role.
00:0000:00
The China World AI Conference
03:16
White House officials often refer to the U.S. and China as being in a race for the latest AI. On September 24, President Donald Trump will meet Chinese leader Xi Jinping. Both governments’ officials have been exchanging barbs over the past few weeks about whether China is stealing AI secrets.
The spokesperson of the Chinese Ministry of Foreign Affairs stated Wednesday that while she had not seen the report in question, China’s AI progress “is the result of its high level of scientific and technical self-reliance.”
At a Beijing regular briefing, MaoNing said: “We believe all parties must strengthen their cooperation in order to promote an open, inclusive and beneficial development of AI that is ethical, for the benefit of humanity as a whole.”
The U.S., she said, should strengthen AI cooperation with China “rather that making baseless accusations.”
Six of China’s leading AI developers were named in the report: DeepSeek (formerly Moonshot AI), Alibaba, MiniMax and StepFun. The report goes into great detail. It includes a list of specific allegations that some Chinese models have derived certain topics from American models, such as “legal specialization,” “agenttic functions,” and “coach/assistant abilities.”
The report stated that “the sheer size of these campaigns, and the sophistication of them, indicate distillation as not only a complement to the AI model development of these companies but is the core.”
Since last year at least, American companies have scattered allegations of distillation. OpenAI claimed that DeepSeek “inappropriately based” its model on ChatGPT. Google claimed in February that it had been bombarded with attempts to copy Gemini, but refused to identify a culprit.
This report includes recommendations to U.S. firms, such as improving the verification of users paying for subscriptions for American-style models. It also suggests that they compare notes about any suspicious behaviors. In certain instances, it recommended that responses to malicious queries be altered or downgraded.
Michael Kratsios (Director of White House Office of Science and Technology Policy) circulated in April a memo to federal agency heads highlighting Chinese distillation of American AI as a crucial issue. The director of the White House Office for Science and Technology Policy, Michael Kratsios, circulated a memorandum to federal agency heads in April highlighting Chinese distillation of American AI systems as a critical issue.
Kratsios announced on X in July that Moonshot’s Kimi K3 was derived from Anthropic Fable, the advanced model.
Kratsios, in a letter that echoed his memo from April, wrote: “Legitimate AI Distillation, used to create smaller and more efficient models, plays an important role in this ecosystem of open innovation.” However, industrial distillation on a large scale, which is done in secret, to steal proprietary U.S. technologies and undermine American research, is not acceptable.

According to a new report, the interest of parents in their children’s education has declined in high-income nations as schools struggle with falling test scores.
Last year, over 760,000 teens in more than a dozen countries participated in the Programme for International Student Assessment. The results released on Tuesday showed that the average score in reading, math and science is at its lowest level in PISA’s history. This includes a continued drop in math and reading scores in the U.S.
China, Singapore and Taiwan scored highest on this test. Estonia, New Zealand and the United Kingdom also scored well.
Students’ surveys may provide the key to understanding why test scores have fallen so dramatically.
In comparison with 2022, nearly every country has seen a decline in the support of families for their children’s education. This is especially true for boys.
Andreas Schleicher is the director for Education and Skills at the Organization for Economic Cooperation and Development (OECD), which administers the PISA exam.
The PISA test scored 35 points more for students who reported that a member of their family asked them about what they had done in class the previous day. This is equivalent to an extra year-and-a-half’s learning. This number is adjusted for differences in socio-economic status.
Another factor is digital distraction. More than a quarter (25%) of high-income students complained that their peers were distracted by digital gadgets in the majority or all science classes. Distraction rates are lower in countries that have strict policies on the use of mobile phones at school.
Schleicher stated that there is a clear correlation between the use of digital devices by students for their leisure time and lower scores on tests. However, when tech is used to learn, results can be more nuanced.
Test scores were higher in countries where students used digital devices between two and three hours a day at school. However, scores dropped once the tech usage surpassed that level.
Asian countries dominate test scores
Asia appears to have figured out how to use technology to the benefit of students. Less than 10% of the students tested in Japan, Korea, and China mainland complained about digital distractions in classrooms. Only four provinces of China, Hong Kong, Macao and Hong Kong participated in the testing.
Asia also had the best schools in the entire world. Japan, Korea and Taiwan were among the 10 top regions for reading, math and science.
In the Chinese provinces, almost all of their children scored above or at baseline in math and reading. It was only 66% in the U.S. for math, and 74% for reading.
The dominance of China in education over the U.S. is a serious concern for experts.
Tracey Brenns, the chief of research and global strategy at the National Center for Education and the Economy (a research group on high-performing educational systems), said: “To win the technological race you will need people who are knowledgeable in science, technology and reading.”
In math, the U.S. scored the same as the average of the OECD nations that gave the test. The U.S. also ranked among the 15 best countries for reading, science and solving computational problems, a brand new category. These rankings were boosted by the declines of other countries. The reading and mathematics scores of U.S. teenagers have fallen to their lowest level since the PISA tests were first given in early 2000s.
The results should be treated with caution, according to researchers. This is because there was not a sufficient number of students and schools participating in the tests.
Less learning, more AI
In general, students with higher test scores were more likely to use artificial intelligence. The students who claimed to never have used chatbots scored higher than those who had. Nearly half of all students from developed countries say that they use AI-chatbots at least once a week to learn.
Schleicher cited China, Japan and Singapore as examples of countries who have successfully used AI in the classroom without it becoming overused. In Asia in particular, he noted, the teachers have more power to design AI systems in the classroom.
In the United States teachers are largely responsible for implementing lessons. He said that in Asia teachers play a major role.
Ink, brushes and calligraphy were used in a classroom where he saw elementary students in China practicing the art of calligraphy. Teachers scanned the students’ work using AI and analyzed how they could improve their handwriting.
Schleicher stated that “the classroom would look like it was in the 1950s if you took a picture — with students writing calligraphy using ink.” But everything was powered with AI.

Stefany Shaheen won the Democratic Primary early on Wednesday morning, in New Hampshire’s 1st Congressional District. This was a highly-publicized race to fill a pivotal swing seat, which attracted a large number of Democratic candidates.
As the incumbent of her district, Rep. Chris Pappas (D-N.H.), seeks a higher office, the Democrats will need to run a crucial midterm race in order for them to gain control of the House of Representatives.
Pappas has been representing the district in Washington since 2019. Last year, he announced his bid to run for U.S. Senate. He was running to replace the older Shaheen.
The Republicans hope to capture the district, for the first in 12 years, while they try to maintain their razor thin House majority.
2026 MIDTERMS STATE OF PLAY: KEY RACES UNFOLD AS STEVENS, EL-SAYED BATTLE IN MICHIGAN
Shaheen narrowly beat Maura Sullivan a Marine Veteran and Obama era Official in a crowded race. According to unofficial results, Carleigh Beriont finished third. She is a progressive and union organizer.
Christian Urrutia is an Army National Guard Officer and attorney in international law. Urrutia is another progressive candidate who ended his campaign this weekend, and gave Shaheen a last-minute endorsement.
Shaheen was endorsed in the final days before the primary election by Carol Shea Porter, a long-time Democratic Rep. who succeeded Pappas. Shea Porter was a Portsmouth City Council member, former Chair of the Police Commission and co-founder of a chronic-care management company.
In recent weeks the primary fight became more bitter. Shaheen’s ad accused Sullivan of having been funded by AIPAC and corporate lobbyists. Sullivan’s ad called Shaheen a ‘fancy lastname, and accused her of being entitled to the support of New Hampshire voters.
TAKEAWAYS #1: The DEMS LEFT-CENTER FIGHT INTENSIFIES as SCHUMER takes another hit; TRUMP’S DECISION IS SPLIT
The daughter of Sen. Shaheen who retired in March announced by her.
She said, “I am so proud of Stefany for answering the call and serving the district where she was raised and born. Stefany has worked in the Portsmouth City Council, and she is a strong advocate for clean and affordable water and health care.”
She’s working to win voters’ approval in her right, and on her merits – and I could not be more proud to support her campaign.
When it came to family, the younger Shaheen was careful to tread carefully. She broke from her mother when she learned of her senator’s agreement to end government shutdown last year. The family name did help her gain support and increase fundraising.
Shaheen’s congressional campaign emphasized such priorities as Medicare for all, tax incentives for first-time homeowners, and overhauling Immigration and Customs Enforcement.
The American Dream now comes with a ZIP Code: Here’s where starter homes are still within reach
Sullivan, on her part, positioned herself as a veteran with experience. She highlighted on her website her service as a Marine, her time as an Assistant Secretary at Veterans Affairs in 2014 under the Obama Administration, and her experience as a health technology executive in the private sector.
I joined the Obama Administration. In a video campaign, Sullivan stated that he was “proud to have worked on national security issues and healthcare reforms to assist other military families.”
Sullivan’s platform calls for cracking down what she called price gouging. It also includes bolstering VA, fighting Trump’s tariffs, and increasing federal investment in housing.
CLICK HERE TO DOWNLOAD FOX NEWS APP
Shaheen now faces off against Republican Anthony DiLorenzo in the general elections.
DiLorenzo, an auto executive endorsed by Donald Trump, beat steel executive Hollie noveletsky who showed her Trump support as she failed to win the GOP nomination for a second time.

Alex Gibney goes against the grain in an age when documentary films about celebrities can be light, sweet and as simple as the snack you’re eating while you watch them. The premiere of his new documentary “Musk” at the Venice Film Festival took place on September 8. It shows that there is still a space for investigative documentaries in Hollywood, as long as viewers are willing to endure its almost four-hour running time.
Gibney never shied away from a controversial topic, whether it was Theranos Elizabeth Holmes or Scientology’s John F. Kennedy in “The Inventor”. The Oscar-winning director is uncompromising in “Musk” as he paints a detailed portrait of the man responsible for turning companies such as Tesla and SpaceX to household names.
Since its announcement in 2023, the documentary, which follows Musk from a 20-something Silicon Valley entrepreneur through to his role as President Donald Trump’s right hand man in his second term has sparked Musk’s ire. Gibney replied, “How could you possibly know?” Musk had called it a “hit-piece” about X at the time. Since then, the two have engaged in a heated online debate.
Gibney managed to get Musk on the camera, sort of.
The documentary includes a Musk avatar that has been created using all of the interviews given by the billionaire. Avatar Musk appears throughout the documentary to discuss everything, from his belief that we are all in a virtual world to his desire to go to Mars.
This is a clever use of new technology to make fun of the CEO. It’s not Musk’s only unflattering segment.
Business Insider showed “Musk”, the same day that it was premiered at Venice Film Festival. Gibney has been hard on Musk three times.
Business Insider contacted Musk to ask for a comment about the documentary but received no response.
Tesla’s Autopilot, the self-driving system that Tesla offers to its customers, has been criticized again.
This documentary explores how Musk became involved with Tesla, and then turned the company into a money-making machine once it went public. The documentary also examines the dark side of Tesla Autopilot – a feature of self-driving cars that was at the heart of many lawsuits.
Gibney compares dashcam footage of a fatal 2019 accident in Key Largo Florida, where a Tesla Autopilot-equipped car killed one person and injured another.
A jury found that Tesla partially was responsible for the accident and was liable for compensatory and punitive damages of $243,000,000 after a Russian hacker had been able to access key crash data Tesla said it couldn’t access. Tesla has appealed.
The story was reported by Business Insider and other outlets, but Gibney’s documentary will not improve the perception of Tesla.
Musk, the ‘Godfather’ of AI, is worried that a rigged election could occur.
Trump and Musk both described DOGE’s purpose as an effort to overhaul federal spending and reform the government. Gibney, a tech reporter from The Wall Street Journal and Wired magazine, says that Musk had access to all Americans’ personal data during his time on DOGE.
Geoffrey Hinton, the “Godfather of AI” and a friend-turned-frenemy of Musk’s, says he thinks Musk is up to something nefarious. In the film, he tells Gibney that he thinks Musk will corrupt the midterm elections with this information.
Hinton writes in “Musk” that “if you want to corrupt midterm elections, and you’re going to use AI to do it, you only need to get detailed data on US citizens.” “What DOGE did looked exactly like what you would have done,” Hinton says in “Musk.”
Gibney, in his voiceover questioning Hinton’s statement: “Is Musk trying to control or connect us?”
Former flames reveal the NDA conditions she refused and Musk’s desire to create a “legion of kids”
It also explores Musk’s private life and his 14 children who are known to the public. The documentary features Justine Musk, Musk’s former wife. However, it is Ashley St. Clair who has given birth to one son in 2024.
St. Clair states in the document that Musk followed her on Facebook and sent her messages telling her to have more kids. St. Clair claims in the document that after their relationship ended and she gave birth to Musk’s child, Musk offered her $15 million upfront and $100,000 a month for 21 years as long she would sign an NDA. She declined.
St. Clair recalled Musk listening to Kublai Khan’s podcast one evening while they were in bed. Kublai Khan is a descendant from Genghis Khan with at least 47 offspring.
Musk also texted to her once that he would need a “legion” of children prior to the Civil War.
St. Clair says Musk didn’t reach out only to her on social media.
She told Gibney that it was a “Gatling Gun of sperm” on Gibney’s DMs.
Musk didn’t respond to an inquiry for a comment about the documentary. Musk responded with “Exactly” on Tuesday when Katie Miller (the former communications director of DOGE) called the documentary “fictional.”
The movie “Musk”, which opens on October 16, will be released in cinemas.