Notizie — Pagina10

Anthropic has launched message threads in public beta, a feature that lets developers building AI agents on Claude send a full conversation once and then only new messages after that. The company says the change cuts the wait before Claude starts replying by about 35 percent on very long requests of 500,000 to 1 million tokens. Prices and rate limits stay the same, and the feature works across all Claude models on Anthropic's platform and through AWS, Microsoft Foundry and Google Cloud. Leggi di più

As Anthropic and OpenAI prepare for initial public offerings that could bring fortunes to San Francisco, more than four dozen laid-off tech workers spent a Monday morning hiking Mount Tamalpais, just across the Golden Gate Bridge. The walk, led by former Google employee Basem Istanbouli, offered a picture of the other side of the AI boom: people who lost their jobs in tech while the industry's biggest names prepare to go public. Leggi di più

OpenAI has released GPT-Rosalind-6, a reasoning model built for biology, drug discovery and translational medicine, but it will only be available to approved scientists, research institutes, companies and government partners. The model uses the same weights as the company's flagship GPT-6 Astra and refuses fewer dual-use biology questions, though it still declines requests that would meaningfully help with biological weaponization. OpenAI rates the model as High capability in biological and chemical risk, below its Critical threshold. The company's system card addendum outlines benchmark gains over its predecessor, the access controls it has built around the model, and the safety tests it ran before launch. Leggi di più

The New York Times sued Brave Software on Monday, accusing the maker of the Brave web browser of scraping its journalism and selling access to it to AI companies. Hours later, the publishers of the New York Daily News, the Chicago Tribune and the Orlando Sentinel filed a separate suit making similar claims. Both cases were filed in federal court in Manhattan and target Brave's search index and its Paywall blocker feature. The complaints say Brave has copied copyrighted articles without permission and let users read paywalled content without paying. Brave did not immediately respond to a request for comment. Leggi di più

Security experts say North Korean hackers may be using artificial intelligence to plan and carry out cryptocurrency thefts, pointing to the growing complexity of recent attacks, including a record theft of more than $6 billion from Binance. The FBI has blamed North Korea for the Binance theft, which it attributes to a group known as TraderTraitor. The use of AI has not been confirmed. Researchers say the tools could make targeted attacks easier to organize and harder to detect. The warnings come as North Korea-linked groups account for a large share of crypto hack losses this year. Leggi di più

AI chatbots such as Claude, ChatGPT and Gemini can search the web for users, and for many questions they return useful results. A traditional search engine still produces better results in some cases. Knowing which tool fits a given question can save time, and the choice is becoming more important as Google pushes AI Mode in its own search engine. Leggi di più

Elon Musk says SpaceX is adding two more sites to its Colossus 2 AI computing complex in Memphis, one he calls Gun Range and another he calls Spam. Musk announced the update on X on Sunday. He said the Gun Range site is on track to be operational by the end of November and the Spam site in December. The expansion comes as residents near the campuses have pushed back through public hearings, protests and petitions, and as the NAACP's lawsuit over unpermitted gas turbines at the facility moves forward. Leggi di più

OpenAI is preparing to combine the Chat and Work tabs in the ChatGPT desktop app into a single experience, according to an email sent to users. The change will first be tested with select users for feedback before a wider rollout. Codex will stay a separate experience, and ChatGPT and Codex will draw from the same token allowance pool. OpenAI said most users won't be affected, though some may need to wait for usage resets more often depending on their usage and model. Leggi di più

A WIRED investigation found roughly 400 apps and websites linked to Chinese-language developers that pass as ordinary AI tools but contain secret features for creating violent AI-generated images and video. The tools include a self-described "pet portrait" app, Viefur, which charges users in digital coins for transforming pet photos into costumed scenes. The reporting shows how easily harmful image generators can hide behind harmless marketing. Leggi di più

New York startup Clark has raised an $11 million seed round led by Touring Capital to build artificial intelligence agents that handle audit work. The company says its agents, called Clerks, review documents, run tests and prepare workpapers much like junior auditors, while human auditors still make the judgment calls and sign the final opinion. Clark says more than 35 accounting firms use its tools, and it offers a 100% accuracy guarantee. The company, formerly known as uiAgent, says it has raised $16 million in total. Leggi di più

Abu Dhabi is using OpenAI's models to help officials check permit and certificate applications against government rules, the emirate's media office said Monday. The pilot, already running on food imports, is one of four OpenAI deals signed that day after Crown Prince Sheikh Khaled bin Mohamed bin Zayed met OpenAI chief executive Sam Altman. The other agreements cover investment analysis through Mubadala, a five-year plan to bring OpenAI's models to government, businesses and schools in the UAE, and an energy deal with state oil company ADNOC. Altman also spoke Monday at the Adipec oil and gas conference in Abu Dhabi. Leggi di più

Anthropic has added a new instruction to Claude Code that tells its models not to recreate specific existing video games, even if the copies are renamed or called original. A social media user first spotted the change, and an Anthropic software engineer confirmed it, saying the aim is to stop users from infringing intellectual property. The rule still allows original games, genre games and games built from a developer's own designs. Still, observers say the move is unlikely to slow a broader wave of AI-assisted game copying, because users can simply switch to other models. Leggi di più

Elon Musk says SpaceX’s Grok Bot cost too much to run relative to the revenue it brings in. He made the comment on X after subscribers reported that their weekly usage limits ran out days early, some of them overnight. Musk did not say how much the agent costs to operate or whether the company has cut subscriber allowances. The posts came less than two weeks after SpaceX’s third-quarter earnings call, where the company touted Grok Bot’s rapid growth. Leggi di più

Israeli startup 42AI has come out of stealth with $62.5 million in seed funding for what it calls a “behavioral world model,” an AI system designed to estimate how people will make decisions in new situations. The company is running the technology at 10 large brokerages, where it helps manage liquidity, margin requirements and hedging exposure. Its founders say the model tracks psychological tendencies rather than past actions, but they acknowledged that the system has limits and that its ability to head off misuse by brokerages' own customers remains an open question. Leggi di più

Google's Gemini AI can summarize your recent emails, draft replies and build to-do lists from your Gmail inbox, and many of those features are turned on by default. Google says it does not train its foundational AI models on personal emails and that Gemini holds onto nothing after completing a task. But the company's privacy promises have exceptions, especially once Workspace content apps are linked to Search services. Users who want to shut the features off can do so through their account settings, though doing so may also disable spelling and grammar checks. Leggi di più

German Chancellor Friedrich Merz said Monday that Germany and Europe must move faster to build their own artificial intelligence technology or risk becoming dependent on outside providers. Speaking at an AI conference in Berlin, Merz called digital sovereignty “a truly existential question” and said Germany must avoid becoming entirely dependent on third parties. He acknowledged the effort will be enormous. The remarks come as the Ifo economics institute finds Germany still poorly positioned to benefit from the AI boom, and as companies including Amazon and Nvidia have joined investment pledges worth more than €225 billion. Leggi di più

A string of comments from tech leaders praising AI for everyday tasks, including parenting and dating, has drawn fresh criticism. On TechCrunch's Equity podcast, reporter Amanda Silberling argued that AI executives may be too far removed from regular people to build products that appeal to them. The debate comes as Anthropic prepares to go public at a time when Wall Street is growing less willing to fund heavy AI spending. Hosts Kirsten Korosec and Sean O'Kane and co-host Anthony Ha discussed whether the disconnect is a marketing problem, a product problem, or both. Leggi di più

A federal judge has refused to dismiss a lawsuit alleging that xAI's data center campus in Memphis, Tennessee, runs gas turbines without proper permits and violates the Clean Air Act. U.S. District Judge Debra M. Brown ruled that the power plant supplying the data center is not mobile, so the turbines are not exempt from permitting requirements. The NAACP brought the suit in April. The company, now known as SpaceXSI, had asked the court to dismiss it, with support from the Justice Department. Leggi di più

China's Ministry of Commerce and Ministry of Science and Technology have proposed adding 40 technologies to the country's export restriction catalogue, 23 of them in artificial intelligence. The draft would control frontier model weights, AI agent architectures and AI-enabled offensive cyber capabilities, with specific technical thresholds. The catalogue changes are open for public comment until November 15, 2026. A separate interim registration requirement for technology exports is already in effect, and it covers exports that have started but are not yet complete. Companies with China-linked AI development or self-hosted Chinese models face the most immediate exposure. Leggi di più

Abu Dhabi's state oil company ADNOC has signed a multi-year agreement with OpenAI and G42 to bring AI agents into its operations. The deal makes ADNOC the first energy company to use OpenAI Frontier, a platform for building and managing AI agents. The first application will focus on planning oil and gas field development, a process that currently takes months and could be completed in hours under the new system. The agreement was signed in Abu Dhabi during a visit to the UAE by OpenAI CEO Sam Altman, and it was announced during ADIPEC, the country's largest oil and gas conference. The companies have not disclosed the deal's value or a rollout timeline. Leggi di più