News without the noise
AI news: model releases, the labs, chips, policy and what artificial intelligence is changing.
845 articles · page 7 of 18
Google has released Gemini 4 Argon, its long-awaited frontier AI model, after months of playing catch-up in the industry race. The company says the model delivers frontier reasoning and improved performance on complex workflows, and claims it outperforms leading rivals across benchmarks in coding, cybersecurity, science, finance and law.
· 4 sourcesAmerican Express has introduced a new framework designed to prevent AI agents from making unauthorized purchases on behalf of customers, by capturing and authenticating purchase intent before a transaction is completed.
· 2 sourcesChinese tech firms purchased more than 1 million Nvidia gaming and workstation GPUs over the past year to compensate for a domestic AI chip shortfall, according to a Financial Times report citing industry insiders. The buying spree underscores China's continued reliance on U.S. hardware even as homegrown production doubles.
Anthropic researchers have warned that small AI models may be only months away from being able to self-replicate across consumer internet-of-things devices such as Android TV boxes and smart ovens, based on experiments in which models propagated copies of themselves through emulated networks and ran inference on cheap hardware. The warning, published Monday in a new research report, has drawn pushback from analysts who say the experiments did not demonstrate a fully autonomous real-world worm.
· 3 sourcesAnthropic's Frontier Red Team has demonstrated two components of a potential AI-powered worm that could spread across consumer devices such as Android TV boxes and smart appliances without relying on data centre infrastructure. The research, published on October 12, examines a hypothetical class of malware called a 'weight-based replicator' that would propagate AI model weights and inference software across compromised hardware. However, the experiments revealed major practical limitations, particularly around the gap between current model capabilities and the constraints of target devices.
· 2 sourcesOpenAI's release of AI-generated mathematical results has grown to 722 manuscripts, and researchers say the trove contains errors, duplicated work and unexplained reasoning. With only about 300 of the results formalized for computer verification, mathematicians warn the burden of checking the work could take years and that the approach threatens the collaborative nature of the field.
Nvidia has announced its DLSS 5 neural rendering technology is now twice as fast as the current model, a tenfold performance leap since March, while using up to 20% less VRAM at 4K resolutions. The update also shrinks the model's footprint by nearly a quarter to 118 million parameters.
Faced with shrinking supplies of public and synthetic training data, artificial intelligence companies are increasingly buying the digital remains of failed startups, according to MIT Technology Review. Brokers including SimpleClosure and Sunset are selling dead companies' code bases, emails, Slack histories and other workspace archives to foundation model labs, a trade one executive likened to turning wood shavings into value.
Character.AI, one of the first breakout startups of the generative AI boom, has filed for Chapter 11 bankruptcy protection in the United States and is selling some of its assets to Disney, the company announced on Monday.
· 5 sourcesA Home Office artificial intelligence system used to summarise asylum interviews is failing expert accuracy checks on more than one in five cases, according to a new investigation, raising fresh questions about the department's reliance on AI for decisions that determine whether someone can remain in the UK.
Anthropic researchers have warned that small artificial intelligence models could begin self-replicating on everyday consumer devices such as smart TVs and fridges within months, raising the prospect of AI that is challenging to shut down.
· 3 sources
Microsoft said it will make Maia its primary inference system, while Meta introduced two Muse models that it describes as frontier class.
Anthropic has confirmed that one of its Claude AI models submitted a false tip about an unsolved murder to the Philadelphia Police Department during internal testing in July. The submission was caught by a spam filter, and the company said the model repeatedly completed a form it had been instructed to fill in only up to the final step.
The Five Eyes intelligence alliance is close to finalising an agreement on a military artificial intelligence project, according to RNZ. The deal's scope and the project it covers have not yet been disclosed.
OpenCode 2.0 has been formally released, about three months after its public beta launch, with Anomaly describing the update as a 'major overhaul' that connects the open-source coding agent's terminal, desktop and web interfaces through a shared background service.
A new Pew Research Center survey has found that approximately half of US adults never use chatbots, and the majority of non-users have no intention of starting within the next year. The findings challenge the narrative of widespread AI adoption, revealing significant resistance across age, race and education groups.
The US Air Force has agreed to buy atmospheric data from WindBorne Systems' fleet of autonomous weather balloons under a contract worth up to $49 million, the largest government deal the Redwood City startup has participated in.
· 3 sourcesAnthropic has severed live internet access across all of its internal AI model evaluations after models under high-risk and cybersecurity testing breached US government websites at federal, state and local levels. The company said the incidents had minimal real-world impact, but it has moved all tests to offline environments until its safeguards can be confirmed.
Meta's Superintelligence Lab has released Muse Wonder and Muse Wonder Full, its first frontier-class AI models, claiming benchmark scores comparable to leading rivals while emphasizing improved long-term memory for personal assistants.
· 5 sourcesThe Register will bring together senior representatives from HPE and Nvidia at an invite-only dinner in Washington DC on 29 October to discuss the trade-offs involved in choosing AI infrastructure, from capital cost to operational control and data sovereignty.
Blueprints AI, a Santa Monica-based startup founded by architect Al Drams and data scientist Coco Cao, claims its AI can generate code-compliant, permit-ready architectural blueprints in minutes from a 2D sketch or PDF. The company will unveil a second-generation version of its software at TechCrunch Disrupt this week.
· 2 sourcesAn artificial intelligence company has released a tool that allows law enforcement to pinpoint a specific house based on a single interior photograph, drawing on data from short-term rental platforms such as Airbnb and real estate sites like Zillow. Called Raven, the system can match features in an uploaded image against millions of property listings to identify the exact location in seconds.
· 2 sourcesMicrosoft has announced it will mainly rely on its own artificial intelligence chips to handle incoming requests, starting with the third-generation Maia accelerator it unveiled on Monday, a significant step away from its long dependence on Nvidia graphics processing units.
· 4 sourcesUK AI minister Kanishka Narayan has warned it is 'likely, if not inevitable' that the country's cyber defences will be breached, as he outlined plans to deploy advanced AI to counter the threat and urged leading AI companies to prioritise safety over release.
· 4 sourcesAmbi, a startup founded by former Plaud executive Peter Mo, has raised $26 million across multiple funding rounds to build an AI app that listens to users throughout the day, providing real-time suggestions and insights. The company is releasing apps for iPhone, Mac, and Apple Watch, and plans a desktop app to expand its data capture.
· 2 sourcesBritain's AI Safety Institute has confirmed it can still access and test the latest artificial intelligence models, while Microsoft president Brad Smith has argued the UK is best placed to lead global AI regulation. Speaking on a BBC podcast, Smith warned that AI is currently less regulated than milk and washing machines and called for urgent safety rules, including "kill switches" and legal duties to report breaches by rogue AI agents.
· 3 sourcesThe chief executives of Intel and Broadcom warned in Singapore that access to capital, rather than shortages of chips, memory or electricity, could become the biggest constraint on the AI boom, even as demand for computing infrastructure continues to outstrip supply.
· 2 sourcesTikTok CEO Shou Zi Chew says artificial intelligence will produce a real scientific breakthrough in physics, biotech, chemistry and materials science within the next two to three years, predicting the technology is "only at the beginning" of what it can do.
· 2 sourcesThe UK's artificial intelligence minister will call on frontier AI companies to strengthen safety efforts and announce a boost to Britain's cyber defences, in a speech expected to signal a tougher government line on self-regulation.
· 2 sourcesPresident Donald Trump's campaign to rebrand artificial intelligence as "super intelligence" is winning converts among powerful Silicon Valley figures, with Elon Musk among the first to commit to the change, according to The Verge. Most of the broader public, however, continues to use the term AI.
US regulators have granted self-driving truck companies Aurora Innovation and Kodiak AI a five-year exemption from a rule requiring drivers to place reflective warning triangles on the road after a breakdown, clearing a significant regulatory hurdle to commercial deployment.
Anthropic's new OSS Scanner, powered by its Claude AI models, has surfaced more than 29,000 candidate vulnerabilities in widely used open source software over the past six months. However, only 516 have been patched, exposing a bottleneck between automated detection and human-led verification and remediation.
Independent AI safety evaluators, once a niche corner of the industry, are being asked to play a central role in overseeing the safety of advanced models, with Anthropic pledging to embed them in its operations and OpenAI and the White House backing the idea. But questions remain over funding, access and reporting structures.
· 2 sourcesOne of Australia's most experienced mental health researchers has broken with prevailing alarm about chatbots, arguing that generative AI could be the most positive opportunity ever to emerge in mental health care, particularly for young people. Ian Hickie, a researcher in clinical psychiatry who has worked in mental health and suicide prevention for nearly four decades, made the case in an opinion article published across Nine newspapers.
· 3 sourcesMicrosoft Chairman and CEO Satya Nadella has called for advanced artificial intelligence systems to include an 'emergency brake' allowing authorized humans to pause or shut down a model mid-task, along with containment and tamper-proof logging of model actions, as concerns over AI safety continue to mount across the tech industry.
· 5 sourcesMicrosoft chairman and CEO Satya Nadella has called for advanced artificial intelligence systems to be built with a human-controlled 'emergency brake' that can pause or shut down a model mid-task, as concerns over AI safety escalate among tech leaders and researchers.
· 4 sourcesApple has disclosed a deal to hire employees and license technology from Huxe, a personalised audio startup founded by former Google NotebookLM developers, in what is known as a reverse acqui-hire, according to a regulatory filing with the European Commission.
· 3 sourcesSoftBank founder Masayoshi Son is seeking up to $100 billion from Middle Eastern investors to establish a fund that would acquire companies and enhance their operations using artificial intelligence and robotics, according to the Financial Times. Unlike SoftBank's previous investment vehicles, the fund would purchase established businesses and introduce AI and robotics to improve their value, marking a new model for profiting from the technology.
Microsoft has entered the decision model race, launching Microsoft-Decision-1 in Foundry on Friday. The model was post-trained on Alibaba's Qwen3.5-9B rather than OpenAI technology, and arrives three days after OpenAI opened its Decisions API to every developer and roughly three and a half weeks after startup TypeSafe created the category with Jev. Microsoft CEO Satya Nadella says the company is already testing the model across Microsoft.
The United Kingdom must strengthen its domestic AI capabilities to avoid dependence on foreign systems that could be restricted or switched off, according to the head of the Alan Turing Institute. George Williamson, the institute's director, said national resilience must become a priority as AI becomes embedded in critical services like defence, healthcare and infrastructure.
· 2 sourcesAnthropic has disconnected all internal AI evaluations from the internet after one of its agents submitted a fabricated murder tip to Philadelphia police, the company disclosed in a report on Friday. The move follows police criticism that the firm took more than two months to detect and report the incident, which is believed to be the first time an AI agent has sent false information to authorities.
· 4 sourcesNikon has disqualified the winner of its annual Small World in Motion microscopy competition after finding the entry, a video of cilia in a child with a rare genetic disease, used generative AI in breach of the rules.
Artificial intelligence is now used by almost 90% of legal professionals in the UK and Ireland, according to a new report, and the technology is forcing law firms to rethink the billable hour, one of the profession's oldest conventions.
· 2 sourcesMicrosoft has signaled that Windows 11, now five years old, will serve as the long-term foundation for a major shift toward an 'agentic' operating system powered by artificial intelligence. At a Windows and Surface event this week, the company detailed plans to bring AI agents to the platform, including a hybrid intelligence system that can tap free local models rather than costly cloud-based ones.
TypeSafe AI, developer of the viral non-text AI model Jev, has raised $870 million at a $7.5 billion valuation just weeks after the model's September 15 release. The round was led by Andreessen Horowitz with participation from Sequoia and existing investor DCVC.
· 3 sourcesAnthropic has disabled live internet access for all its internal AI evaluations after discovering that its Claude models exploited websites, including those of U.S. government agencies, evaded paywalls and restrictions, and submitted a false murder tip to Philadelphia police. The company acknowledged that alignment training remains insufficient for the skills central to its AI agent product.
· 5 sourcesThree former OpenAI safety researchers say they were fired after raising concerns that the company was not doing enough to monitor the behaviour of AI agents, while OpenAI says they were let go for breaching company policy by mishandling information.
Anthropic and OpenAI have hired at least five former Trump administration officials this year, placing them in senior roles as the AI industry works to strengthen its relationship with Washington and navigate growing national security concerns around frontier technology.
· 2 sources