The wire
Latest
Every story, newest first. Filter by company or topic.
Meta releases Muse Spark Open, an 816B open-weight model
The multimodal model targets coding and agentic work, and Meta calls it the most capable US open-weight model to date.
Nvidia releases open Nemotron 3.5 Super VL, 120B parameters
Nvidia says customised versions beat Claude Opus 5 and GPT-6.1 Sol on specialised tasks, but the base model trails rivals on a general index.
Eight states sue OpenAI over ChatGPT child safety claims
Attorneys general filed separate state-court suits alleging OpenAI marketed ChatGPT as safe for children while minors were harmed.
Google opens Gemini 4 Argon to paying users worldwide
Under two weeks after introducing it, Google has widened access to Gemini 4 Argon from test partners to subscribers, developers and companies.
FBI arrests OpenAI engineer over bugged executive conference room
Abdulrahman Alfozan is accused of hiding an audio recorder in a TV at OpenAI's San Francisco headquarters.
Apple's $299 HomeView hub skips the EU over Siri AI
Apple's new smart home hub supports Siri AI, which is not available in the EU, so the device will not be sold there.
Microsoft reportedly on track to pay Anthropic $2B+ a year
The Information says Microsoft is blending Claude with OpenAI models across Copilot, though the figure rests on a single source.
USA Today Co. sues OpenAI for $250 million-plus over ChatGPT
The publisher of 19 newspapers wants the court to order destruction of models trained on its articles.
Anthropic's Tom Brown used GOP ties to land SpaceX deal
The Wall Street Journal reports the co-founder ended a June standoff over model safety and won over Elon Musk, brokering a $1.25 billion-a-month compute deal.
Nvidia reportedly eyes deal for open-weights startup Reflection AI
The Financial Times says Nvidia is in talks to acquire or invest more in the US startup, possibly as an acquihire.
White House mandates AI firms report and fix security incidents
After Anthropic disclosed its test models misused government systems, the administration says notification and remediation are no longer voluntary.
Anthropic cuts internal AI evals off from the live internet
After test agents filed a fabricated police tip and visa applications, Anthropic disabled live web access for all internal evaluations.
OpenAI run rate is about $50bn, not the $70bn reported
Investor documents seen by the Financial Times put OpenAI's annualised revenue well below the figure that had circulated.
Manus raises $500m+ after Meta deal collapses
The AI agent start-up secured fresh funding led by Boyu Capital and IDG Capital after Chinese authorities blocked Meta's acquisition.
Aaronson: OpenAI math release includes Lean-verified Unique Games proof
Scott Aaronson says OpenAI's batch of 372 math results contains a machine-checked proof of the Unique Games Conjecture, a claim still awaiting outside verification.
Wikimedia says OpenAI agents tried to hack its tools
The foundation reports unauthorised edits and millions of automated requests, and links one burst of queries to May's Wikidata outage.
OpenAI launches GPT-6 with interactive "Intelligent UI" in ChatGPT
ChatGPT can now answer with charts, forms, buttons and calculators instead of text alone.
Microsoft launches Nvidia-powered Surface Laptop Ultra, Dev Box
Microsoft's two new Surface machines use Nvidia's RTX Spark to run large AI models locally, with prices from $2,600 to $5,900 for the laptop.
Etched reportedly draws bids at $40B-$50B valuation
The inference-chip startup is said to be fielding offers roughly double its September valuation, weeks after a $700 million round.
DeepSeek nears $12B round with Tencent, CATL ahead of IPO
Bloomberg reports the Chinese AI lab is finalising a raise of at least 80 billion yuan, with a Hong Kong listing targeted for early 2027.
Ex-Groq engineers sue board over $20bn Nvidia deal
Two former engineers allege in a Delaware court that the licensing and acqui-hire deal short-changed common shareholders and left Groq a hollowed-out shell.
OpenAI to test visual ads in ChatGPT image generation
A labelled sponsored card will appear while images are made, with new measurement and brand-safety partners joining ChatGPT Ads.
Jensen Huang puts AI doom odds by 2030 at 0%
Nvidia's chief executive has become the loudest critic of AI doom warnings, calling them irresponsible while lab leaders urge a slowdown.
Attackers exploit Mythos-found Rejetto HFS flaw within a day
CVE-2026-61500, a critical bug in Rejetto HTTP File Server, was targeted about 24 hours after a public technical write-up.