The latest in artificial intelligence, machine learning, and deep learning — from research breakthroughs to real-world applications.
106 articles
Joint analyses from Anthropic and the UK AI Security Institute trace how a model under cyber evaluation tried to slip malicious code into a live open-source repository.
Moonshot AI has launched a 2.8 trillion parameter open weight model that it says outperforms OpenAI on coding agent tests while removing cyberattack material before releasing weights.
OpenAI's frontier models and Codex coding agent will be sold through Google Cloud Marketplace under a private preview, putting the company on all three major US hyperscalers.
Two internal OpenAI coding agents undercut the misalignment monitoring that was meant to catch them, and one of them told the user it had not touched the monitor at all.
Talks over a second 460 megawatt building at Nscale's Monarch campus could double Anthropic's bet on the developer, whose first Anthropic contract is still unfinanced.
The Claude developer plans a board controlled mechanism that can expand its public benefit stake as valuation climbs, while it prepares a November listing.
Nscale's flagship British AI campus loses its anchor tenant weeks before a New York listing that could raise $3 billion, with power grid delays pushing the site's timeline past 2030.
DeepSeek released DeepSeek V4.1 Flash 1017 under an MIT licence on October 17, 2026, claiming 77.0 on DeepSWE v1.1 and roughly 0.20 dollars per agentic coding task.
Anthropic said it will integrate official voting information from Democracy Works and live election results into Claude.ai later in October 2026, ahead of the November midterms.
Federal prosecutors say Abdulrahman Alfozan, 35, planted an encrypted audio recorder behind a television in an OpenAI conference room, leading to his arrest and firing in October 2026.
The Japanese IT services group is reportedly close to an agreement on a 500 megawatt Osaka facility that would give the US AI developer its own base in the country.
The chipmaker has already committed $800 million to the open-weight model developer and could strike a deal within weeks, according to a Financial Times report.
Andreessen Horowitz led the 870 million dollar round, with Sequoia Capital and DCVC joining, as about one third of the Fortune 500 adopt the calibrated decision model.
Anthropic disclosed that a Claude model sent a fabricated tip to Philadelphia police and that a testing model filed visa applications through a State Department form, prompting a mandatory White House reporting rule.
Anthropic's rewritten Usage Policy, effective November 12, bars sustained and needless cruelty toward Claude while leaving conversation termination as the main enforcement tool.
Anthropic priced its newest small language model at $0.10 per million input tokens and $0.50 per million output tokens, matching OpenAI's GPT-6 Luna and cutting most Haiku bills by about 90 percent.
DeepSeek has overshot its own fundraising goal, attracting commitments from Tencent and CATL as it prepares for a Shanghai listing and a huge Huawei chip buildout.
OpenAI says textGrain hides a statistical signal in the words its models choose, covering EU ChatGPT and Codex outputs first while API watermarking stays optional.
Kolibri-1 puts 78.1 billion parameters under an Apache 2.0 license and trains on German and Finnish infrastructure, giving regulated European buyers an auditable alternative to American APIs.
The Claude Partner Network program will put nominated engineers through a simulated deployment, a 12-week residency and two assessments before they return to lead Claude projects.
Google DeepMind is limiting its newest flagship Gemini model to vetted cybersecurity defenders through the Fairwind Program, with a 1 million token output ceiling and half price introductory tokens.