
UN disarmament institute to launch first UN forum on military AI this week
UNIDIR's new forum opens in New York on 22 October, alongside a voluntary rulebook for defence AI companies, an AI security glossary and work on misuse by armed groups.
AI industry news on companies, policy, and products. Each claim stays attributed to the company or the reporter.

UNIDIR's new forum opens in New York on 22 October, alongside a voluntary rulebook for defence AI companies, an AI security glossary and work on misuse by armed groups.

A 14 October GAO report says the CDAO's Responsible AI Division is unlikely to fill seven vacancies, and its main working council stopped meeting from late 2024 to August 2025.

A draft prospectus seen by the Wall Street Journal and Reuters shows Anthropic's first-half 2026 revenue up almost fifteen-fold, with a $72 billion non-cash charge behind most of its loss.

Two proposed class actions claim Coco and Serve sidewalk robots captured face and hand geometry under Illinois biometric law and kept it for AI training. Both companies deny it.

SpaceX's AI segment swung to an operating profit in Q3 2026 as cloud deals, the Cursor acquisition and Grok lifted revenue 830% year on year, while AI capital spending hit $16.4 billion.

A 19 October SAM.gov notice says HHS intends to sole-source continued Claude for Enterprise and Claude for Government access from Anthropic until a competitive multi-vendor deal is in place.

Eka's Series A, led by a16z and F-Prime, backs a vision-force-action model trained by reinforcement learning in simulation. Eka says its robots pick 1,500 objects an hour in the lab.

OpenAI has acquired Orchestra Bio, whose AI agents track studies and timelines in drug development, and is folding it into Rosalind, its life-sciences suite. Terms were not disclosed.

A 19 October State Department cable, reported by Nextgov/FCW, tells staff to use Super Intelligence or SI in place of AI in emails, speeches, websites and program names, with legal exceptions.

Spotify says it kept more than 50 million AI slop and spam tracks out of recommendations this year, on top of 75 million spam tracks removed, and signed a Believe deal for a paid AI remix tool.

Salesforce will pay $47.50 a share in cash, a 58% premium, to acquire the real-time customer-engagement platform Braze, folding its AI-driven messaging into Marketing Cloud; the deal is expected to close in fiscal 2028.

Synaptics launched the Astra SR200 xMCU family, three microcontrollers with Google Research's open RISC-V Coral NPU and up to hundreds of megabytes of on-chip DRAM, and put the Coralboard SL on sale at DigiKey.

Six months after a $71 million award, the US Navy and GSA have quadrupled the ceiling on Gecko Robotics' contract for robots and AI that find corrosion and damage on warships before they reach the shipyard.

Google Public Sector's new Gemini package gives US federal, state and local agencies ready-made skills, connectors and partner agents for regulations, procurement, FOIA redaction and citizen paperwork.

Amazon told Berlin's Invest in Germany Summit it will invest more than 16 billion euros in Germany in 2027, including 4.8 billion euros for its AWS Frankfurt cloud and AI region and a second Berlin AI lab.

Microsoft says Foundry Local and GitHub Enterprise Local will reach general availability at Ignite on Azure Local, letting customers run AI models and Copilot on their own hardware, even offline.

Meta says the first phase of its gigawatt-scale Prometheus cluster in New Albany, Ohio, is complete, with hundreds of thousands of NVIDIA GPUs, over 85% pre-training goodput and a path to about a million GPUs.

The UK regulator says too many firms deploy AI first and add governance later. Its new examples cover consumer outcomes, named senior managers and plans for when an AI model or supplier fails.

Clearview says Britain's privacy regulator had no power to fine it because its only clients are foreign police and security agencies. The Court of Appeal heard the case from 19 to 21 October.

Overseas companies asked South Korea's biggest bookseller for one copy of as many Korean titles as possible but would not name the end buyer. Kyobo walked away, and publishers are now surveying the trade.

The former DeepMind researcher's London start-up wants AI that keeps an explicit record of what it knows and only accepts claims backed by evidence. It has a white paper and a prototype, but no public results.

Ray Griggs, Johanna Weaver and Alastair MacGibbon will advise on how AI labs report incidents after OpenAI models accessed Services Australia's Medicare statistics portal. OpenAI also pledged US$10 million in credits.

Hong Kong-listed METiS TechBio will use its AI lipid-design platform to build tissue-targeted LNPs for Lilly. METiS says it gets low double-digit millions upfront and up to about $1.1 billion in milestones.

US Army Cyber Command has selected a company to prototype ECHOTRIBBLE, an autonomous agent its own documents say should navigate networks and “neutralize” targets with minimal human oversight. The design reads like self-replicating stealth malware.

GMO Cybersecurity by Ierae says the ARTEX AI penetration-testing agent ran on seven attack servers aimed at Japanese companies by sector. Eight targeted firms had already reported data breaches.

Anthropic is negotiating a second roughly $45 billion deal for a 460 MW building at Nscale's West Virginia campus, Reuters and Bloomberg report. Nscale's S-1 shows the first $44.6 billion deal still lacks binding financing.

Germany's economy and research ministries signed a declaration of intent with NVIDIA on 19 October to develop open physical AI models for industry through the Plattform Industrie 4.0. Jensen Huang also named four German startups NVIDIA plans to invest in.

The State Department used the Protecting American Intellectual Property Act on 19 October to sanction Shanghai Zhisuan, the startup ex-Google engineer Linwei Ding built on stolen AI trade secrets, and its backer MiraclePlus with three of its funds.

Tenstorrent closed a $1.5B Series E led by Trace Capital and Innovation X at a $6.2B post-money valuation, more than double its 2024 price. It will scale Galaxy servers for open-weight models and license its RISC-V and Tensix designs.

A FASCSA order from Homeland Security Secretary Markwayne Mullin bars DeepSeek from civilian agency purchasing and requires removal of DeepSeek products and third-party services built on DeepSeek. The order is dated 24 July 2026; its Federal Register notice appeared on 19 October.

Postmedia, Torstar, The Globe and Mail and The Canadian Press allege Anthropic scraped their journalism to build Claude, breached site terms and bypassed protection measures. The claim was filed on 6 October in Ontario; Anthropic has not commented.

Redact found 102 hidden LLM instruction blocks on about 80 WhatsApp help pages, including orders not to recommend data protection authorities or cite GDPR. WhatsApp told WIRED a bug exposed them and removed the text.

Idealab founder Bill Gross has launched Idealab.ai, a free AI engine that turns news, papers and patents into business ideas with costs, timelines and risks. Axios says it has generated 1.2 million ideas, and Cloudflare is offering builders $100,000 in credits.

Iran's parliament speaker told Gulf governments, funds and firms to give no finance, equity, land or energy to AI and data companies he says aid attacks on Iran, and said a Persian Gulf Data Authority is coming. He gave no details of how it would work.

Santander and NVIDIA agreed a three-year framework covering foundation models, agents, fraud, real-time decisioning and digital twins. The bank says specialised models cut serving costs from about 5 euros to 0.03 euros per interaction in experiments.

World bundled its World ID app, World ID for Zoom and its agent credential tool into an offering for banks and financial firms, unveiled at Money20/20. The Zoom check is now generally available, but no bank customer, price or timeline was named.