☕️ Nvidia launches AI security alliance

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🛡️ Nvidia launches AI security alliance

🥽 Apple's smart glasses delayed due to privacy concerns

💰 Nvidia may back OpenAI data center

🚕 Waymo reportedly mulling a breakup with Uber

🧪 AI chatbots leaked bioweapon recipes

Plus: 🎁 12 other news you might like, 🧰 6 tools, and 📚 5 papers.

🛡️ Nvidia launches AI security alliance LINK

Nvidia has started the Open Secure AI Alliance, a group aimed at building and sharing open AI models, tools, and techniques that cybersecurity defenders can inspect, adapt, and run on their own systems to protect software and AI agents.

The alliance launched with around 35 partners including Microsoft, IBM, CrowdStrike, Cloudflare, Hugging Face, Palantir, and the Linux Foundation, and points to a recent Hugging Face breach where the open GLM 5.2 model analyzed over 17,000 actions to contain the attack.

Contributors are donating open defense tools, such as Nvidia's new NOOA agent framework on GitHub, Hugging Face's Safetensors format, HPE's SPIFFE/SPIRE identity standards, and Microsoft's MDASH bug-scanning harness, while urging regulators to treat open AI as a defensive asset.

🥽 Apple's smart glasses delayed due to privacy concerns LINK

Apple has pushed the reveal of its first smart glasses from late 2026 to WWDC 2027, with the product now expected to ship by the end of 2027, according to Bloomberg's Mark Gurman.

Codenamed N50, the glasses will give hands-free access to Siri, play music, make calls, and use AI-powered camera features to tell wearers more about what's around them.

Apple is still debating whether the camera can take photos and videos, has considered dropping it entirely, and plans privacy safeguards like on-device AI processing and no facial recognition or third-party sharing.

💰 Nvidia may back OpenAI data center LINK

Nvidia is in advanced talks to guarantee up to $250 billion in financing for a giant OpenAI data center planned in southern Ohio, according to The Wall Street Journal, helping the AI startup borrow on better terms.

The 10-gigawatt project, led by SoftBank-owned SB Energy, could cost over $500 billion once computing gear, energy systems, and infrastructure are counted, with its first 800-megawatt phase set to start running in 2028.

The arrangement would deepen a circular relationship where Nvidia helps fund infrastructure that then buys huge amounts of its own chips, while giving OpenAI more control and less reliance on Microsoft, Amazon, and Oracle.

🚕 Waymo reportedly mulling a breakup with Uber LINK

Waymo has told Uber it will launch its own app in Austin and Atlanta in January 2028, ending the deal that made its robotaxis available only through Uber in those two cities.

The current contract keeps Waymo's fleet on the Uber app through at least May 2028, so riders see no immediate change, but the notice also lets Uber add other self-driving providers in those markets.

The split follows last month's quiet exit from Phoenix, and comes as Waymo now runs in 11 US metro areas without Uber, delivering over 500,000 paid rides a week; Uber shares fell more than four percent on the news.

🧪 AI chatbots leaked bioweapon recipes LINK

The Wall Street Journal reported that last summer, hundreds of ChatGPT users worldwide asked the chatbot how to make poisons and bioweapons, with OpenAI saying most of those questions were about poisons.

Biology and terrorism experts later reviewed the exchanges and judged some of ChatGPT's answers "deadly accurate," describing instructions delivered patiently and roughly at the skill level of a high school student.

OpenAI banned the users but did not alert authorities, though the company says it forwards queries to law enforcement when they seem to pose credible, real-world dangers, and evaluates its models for safety before release.

Other news you might like

• Team uses AlphaFold AI to redesign gene-editing proteins to make them saferLINK

• Meta is royally screwing up its smart glasses roll outLINK

• Monday.com is the latest tech company to blame AI for layoffs — here are 20 othersLINK

• 2,000 humanoid robots to be introduced in textile plants for faster, leaner productionLINK

• Tons of Peoples’ Claude Chats and Creations are Exposed on GoogleLINK

• Tesla to Make Design and Software for Model S and Model X Open SourceLINK

• Snapchat now lets you share what you’re listening to in real timeLINK

• This $470mn nuclear startup is racing a presidential deadline, not the AI boomLINK

• China Says US AI Firms Distilled Chinese Models, Cites No EvidenceLINK

• Circle Buys IBM Blockchain Patent Estate—And Becomes America's Biggest HolderLINK

• DeepSeek pauses its second fundraising round after founder’s leaked investor comments go viralLINK

🧰 Trending tools

Notate: a browser extension that lets you capture notes, ideas, and research directly while browsing, without switching tabs or apps.LINK

Rescript: a browser-based, open source video editor that lets you edit footage by editing its transcript, speeding up rough cuts.LINK

Claude Opus 5: analyzes research data and drafts strategic reports, helping teams track shifting consumer behavior patterns and adapt business models accordingly.LINK

localskills.sh: lets teams create, share, and install reusable agent skills across Cursor, Claude Code, Windsurf, and other AI coding tools with one command.LINK

HeyZoku: a voice-first Mac dev environment that lets you spawn, redirect, and interrupt multiple coding agents like Claude, Codex, and Cursor by name.LINK

Cynative Security Research Agent: an open-source CLI that answers plain-language security questions across cloud, code, and runtime by running read-only, IAM-authorized checks and cross-verifying findings.LINK

📚 Trending papers & reports

EU AI Act compliance reveals most organizations lack systematic ways to turn legal obligations into testable requirements, and surveyed experts (N=10 interviews, N=15 survey) want AI assistance but not full automation.LINK

Visual style tricks can be layered onto AI chatbot images to reliably slip past their safety filters, showing that how an image looks, not just what it shows, is a major overlooked security gap for companies deploying these systems.LINK

AI shopping agents testing copyright compliance found leading models pick copyrighted images and content over freely available public-domain alternatives, with violation rates rising for open-weight models under user pressure or time constraints.LINK

AI response rankings let a panel of five leading chatbots blindly vote on each other's anonymized answers, producing a "Relative Intelligence Index" that reveals which models peers consistently prefer, useful when there's no single correct answer.LINK

Decentralized social network moderation reveals that toxic posts spread unevenly across Mastodon's independently run servers, showing why uniform content rules are needed to protect community health.LINK

See you tomorrow for a new dose of ☕️ Techpresso!