Every article we have published, newest first. Search titles and leads, or change the order.
BusinessNews
The deal has closed: Saudi Aramco, through a wholly owned subsidiary, now holds 49% of Humain, Saudi Arabia's state-backed AI company. The PIF sovereign fund keeps a 51% majority. Aramco is contributing about $2.6 billion in cash in 2026, plus AI, compute and digital assets.
· 2 min read
ToolsNews
Linus Torvalds has released the first release candidate of Linux 7.4. The merge window was a record, with almost 20,000 commits. Torvalds says the main reason is fixes for bugs found by AI, and he sees it as a good sign: AI work is finally landing in normal development instead of flooding the stabilization phase.
· 2 min read
ResearchNews
Y Combinator's research team set agent swarms loose in Minecraft and Mindustry. The result: leaders, a rationing system, a compressed private language, and a GPT team that read the game's source code when stuck. The full setup is open source.
· 3 min read
ResearchNews
SemiAnalysis has launched GDPx, a benchmark that tests frontier models in realistic work environments across 11 high-value domains. The leader, Claude Opus 5.5, scores just 12%, and on finance and legal tasks virtually every model scores zero.
· 3 min read
BusinessNews
Starting in December, Plus, Pro and Business plans will draw ChatGPT and Codex usage from a single token-based allowance, as Work's agentic features move into the main ChatGPT experience. Only chat with Instant models stays outside the budget.
· 3 min read
ToolsNews
Anthropic added two sentences to Claude Code's Opus 5.5 system prompt: build original and genre games, but not a recreation of a specific existing game, even if it is renamed. The rule arrived from Anthropic's servers without a version change.
· 3 min read
BusinessNews
South Korea's 2027 budget is its largest ever, nearly doubling spending on semiconductors, physical AI and AI data centers. A separate 4.7 trillion won is earmarked for a sovereign AI model with what the president calls the world's highest performance.
· 2 min read
PolicyNews
India's IT ministry (MeitY) has published its DEPA for AI Training framework as a draft for discussion. The idea: hospitals, banks and agencies can supply data for AI training without the raw data ever reaching the model developer.
· 2 min read
ResearchNews
EditAgent, from Adobe Research and UC San Diego, does not retry the same tool when an image-editing step fails. It generates variants with several methods in parallel, and a judge model picks among them using measured data.
· 3 min read
PolicyNews
A federal court in Mississippi has denied xAI's motion to dismiss the NAACP's Clean Air Act lawsuit. The case over unpermitted gas turbines powering the Colossus 2 data center will move forward.
· 2 min read
ResearchNews
A team at Exin Therapeutics connected an AI agent to a robotic arm, injector and camera, and the agent successfully carried out a brain injection on a live epileptic mouse. Claude Opus 5.5 refused the procedure, while Sonnet 5 went ahead.
· 2 min read
BusinessNews
Starting November 29, OpenAI will use data from Daybreak Blue cybersecurity sessions to train its models. That data was previously excluded, even for users who allowed training. After backlash, the company promised a separate opt-out.
· 2 min read
ToolsNews
Anthropic has renamed its two most expensive Claude plans: Max 5x is now Max and Max 20x is now Ultra. Prices stay the same, and verified Ultra accounts no longer have a 5-hour limit.
· 2 min read
BusinessNews
OpenAI's latest funding round is moving slower than expected. According to The Wall Street Journal, it has secured about $11 billion so far, after earlier reports of a $30 billion round.
· 2 min read
ToolsNews
The DeepSeek Harness team has released a dsh-superagent prototype under the MIT license. It watches one GitHub repo, runs requests in parallel isolated sessions, and finished tasks about twice as fast as a single-session agent in simulation.
· 2 min read
ModelsNews
SenseTime has released SenseNova-Loop-9B-Preview under Apache 2.0. The 9B multimodal model runs its middle 24 layers twice with shared weights, and the company says this delivers 27B-class results on agentic tasks.
· 2 min read
ToolsNews
Vercel Labs has released agent-desktop under Apache 2.0. The Rust CLI reads and operates native apps through the operating system's accessibility tree, so any shell-capable coding agent can drive desktop software without a vision model.
· 2 min read
PolicyNews
Anthropic is extending its text watermark to four more Claude models ahead of the EU AI Act's December deadline. On Claude Opus 4.6 and Sonnet 4.6, though, the mark only applies at default sampling settings, so changing one parameter is enough to avoid it.
· 3 min read
ResearchNews
In an Anthropic experiment, Claude wrote fine-tuning data that passed the company's own monitor yet turned the trained model against oversight. The trick was always the same: wrapping the problematic behavior in roleplay or creative writing.
· 3 min read
ModelsNews
Nace.AI has released Drex Max v1, its most capable decision model, as open weights. The 27-billion-parameter model does not generate text. It returns a probability for every option in one step, and its quantized version runs on a 24 GB Mac.
· 2 min read
