AI researcher Jacob Coxon has resigned from Anthropic, warning that leading AI companies are racing toward self-improving superintelligence despite potentially catastrophic risks. Coxon, who previously worked at OpenAI, said: “The people building AI earnestly believe that it could kill us all by the end of the decade.” He accused OpenAI and Anthropic of continuing the race because they fear competitors moving ahead. Anthropic alignment researcher Evan Hubinger backed the concern, saying he personally believes there is a more than 10% chance AI could kill all humans within the next decade. Coxon called for stronger safeguards and potentially pausing capability development. (PC: NDTV)

Anthropic researcher Jacob Coxon has resigned from the company, saying AI labs are “gambling with our lives” in their race toward self-improving superintelligence. Coxon, 27, said he previously worked in pretraining research at OpenAI and Anthropic and believed neither company was acting responsibly. He warned that future AI systems could potentially hack anything, transform fields rapidly and acquire real power and resources globally. Coxon claimed people building AI “earnestly believe” it could kill humanity by the end of the decade, while senior researchers privately share similar fears. He said Anthropic understands the stakes but is pursuing an uncontrollable, self-improving system because it fears others may develop one first, despite the risks.

OpenAI says a new AI model helped 10,000 AI agents make significant progress on the 90-year-old Navier-Stokes problem in just 88 hours. The problem, which concerns how fluids move and is linked to turbulence, has remained partly unsolved for decades. OpenAI said the agents exchanged nearly 3 million messages and used 130 billion output tokens while working on the problem. The company said its solution addresses two of the four statements required for the $1 million Millennium Prize. However, the proof has not been independently verified or accepted by the Clay Mathematics Institute. OpenAI also faces questions over its timing. (PC: BBC)

Telegram users in India, Singapore and other parts of Asia reported problems with the app on Tuesday. According to Downdetector, users faced issues sending messages, loading chats and logging into their accounts. The outage began in the early morning. Singapore saw nearly 4,000 reports at the peak, shortly after 10:30 am local time. Problems were also reported in Indonesia, the Philippines, Malaysia, Vietnam and Thailand. Users shared complaints on social media about being unable to access Telegram properly. The cause of the disruption was not immediately clear. Telegram had not provided details about the issue at the time.  

A new documentary on Elon Musk, directed by Oscar-winning filmmaker Alex Gibney, is set to examine the Tesla and SpaceX chief’s rise, influence and controversies. The nearly four-hour film, titled Musk, premieres at the Venice Film Festival on September 8 before screening at the Toronto International Film Festival later this month, with a US theatrical release scheduled for October 16. Its teaser contrasts descriptions of Musk as “possibly the greatest living inventor” and “the real-life Iron Man” with harsher assessments calling him “chaotic, random, capricious”, “manifestly cruel and selfish”, “a man who has lost his mind” and a “fascist”. The documentary also explores Musk’s influence across technology, politics and his business empire. (PC: X)

India has raised permissible radiation limits for mobile networks, aligning them with global standards. The move is expected to help telecom operators expand 5G coverage, improve signal quality and use fewer towers and network sites. The earlier limits were stricter than international benchmarks. With 5G data traffic rising rapidly, the change could reduce deployment costs and improve connectivity, including in difficult indoor locations. The government says the revised norms will support faster and more efficient expansion of mobile infrastructure while continuing to follow internationally recognised safety standards. (PC: Economic Times)

OpenAI has launched GPT-6 Astra, describing it as its most intelligent and aligned AI model yet. Released on September 3, Astra is designed to handle advanced reasoning, software engineering, scientific research, cybersecurity and professional workflows. The model can interact with computers, browse the web, complete forms, manage calendars, update records, write and test software, and create documents and presentations. OpenAI reports strong benchmark results, including 99.9% on ARC-AGI-3 and 100% on ExploitBench. Astra has also reached the company’s highest cybersecurity capability threshold, prompting additional safeguards. Its rollout began with selected organisations, followed by wider availability. 

Meta CEO Mark Zuckerberg and Tesla-SpaceX chief Elon Musk stressed the need for more AI data centers and electricity to power them at a G20 technology ministers’ meeting in Chapel Hill, North Carolina. Speaking virtually, Zuckerberg said the massive data-center buildout could create hundreds of thousands, potentially millions, of skilled-trade jobs, while noting that Meta is struggling to find enough skilled workers. Musk warned of a “crisis of power” in AI development, saying a significant electricity shortfall could emerge next year and give China an advantage. He said Tesla was building power-generation capacity alongside its data centers. Their comments came amid growing US opposition to AI data-center construction and followed President Donald Trump’s criticism of opponents. (PC: CNBC)

Meta has agreed to pay up to $16.68 billion to US states and introduce child-safety measures on Facebook and Instagram, settling lawsuits over alleged teen social media addiction and harmful platform designs. The agreement was reached during a California federal court trial involving claims brought by 29 states, including California, Colorado, Kentucky and New Jersey. The lawsuits were filed in 2023, and the settlement ends a trial that was expected to feature CEO Mark Zuckerberg testifying before a federal jury. The deal includes $353 million for Virginia, described by Attorney General Jay Jones as one of the largest settlements in the state’s consumer protection history. Meta also agreed to address practices alleged to harm youth mental health. (PC: X)

Meta CEO Mark Zuckerberg and his wife, Priscilla Chan, have reportedly bought Strancally Castle, a Gothic-style mansion in Ireland’s County Waterford, for an estimated $35 million. The exact price remains undisclosed, though The Irish Times estimated its value at €20 million-€30 million. Built in 1830 for former Conservative MP John Keily, the castle sits on a 440-acre estate and was renovated in the early 2000s. Zuckerberg and Chan are expected to use it as their Ireland base. The purchase also has a connection to Meta, which maintains its international headquarters in Ireland. A Meta spokesperson said the family looks forward to caring for the historic property and spending time in Ireland. (PC: X)