
Mistral's 1T cybersecurity model, spy plots, and AI eating the web
Mistral drops a trillion-parameter open-weight security model, Reddit retreats behind a login wall, and a Chinese spy plot targeted Jensen Huang.
Subscribe free →
Mistral Drops a Trillion-Parameter Cybersecurity Model — and Opens the Weights
Mistral released its largest model yet — a trillion-parameter beast specifically tuned for cybersecurity — and made the weights publicly available. The pitch: organizations handling sensitive security operations shouldn't have to trust a closed-model API with their crown jewels, and open weights give them full control over deployment and fine-tuning.
Why it matters: A 1T open-weight model purpose-built for security is a direct shot at proprietary players like Palo Alto and CrowdStrike's AI tools — and gives defenders the same scale advantages that were previously locked behind paywalls.

Reddit Is Walling Off the Open Web Because AI Is Killing Search Traffic
Reddit announced it's moving to a mandatory logged-in experience, ending the era where anyone could browse threads without an account. The driver is blunt: AI-powered search is vacuum-sealing organic traffic that used to flow from Google, and Reddit needs to own its audience directly rather than lease it from search engines.
Why it matters: Reddit's pivot is the clearest signal yet that AI search is materially destroying referral traffic for content-rich sites — and the open web's slow retreat behind authentication walls is just getting started.

The US Investigated a Chinese Spy Plot Targeting Nvidia's Jensen Huang
A senior executive at a Singapore-based AI cloud company — already accused of illegally routing Nvidia chips to China — was also reportedly investigated as a potential Chinese intelligence asset with Huang as a target. The New York Times broke the story, citing government sources familiar with the probe.
Why it matters: Chip smuggling is one thing; running intelligence operations against the CEO of the world's most strategically important semiconductor company is another — this raises the geopolitical stakes of the Nvidia export control saga to a new level entirely.

OpenAI-Backed Thrive Holdings Is Buying ADT for $15.3 Billion
Thrive Holdings, which has OpenAI backing, is acquiring home security giant ADT in a deal that values the company at $15.3 billion. The plan is to layer AI into ADT's monitoring infrastructure — think smarter threat detection, automated response, and personalized security profiles across millions of homes.
Why it matters: This is one of the largest AI-driven acquisitions of a legacy physical-world company to date, and it signals that the race to put AI agents in the home is moving beyond smart speakers into security and safety infrastructure.

Google Gemini Can Now Generate and Edit Docs, Sheets, and PDFs Inline
Google rolled out file creation and editing directly inside Gemini chats — PDFs, Google Docs, Sheets, slide decks, Markdown, and code files can all be generated, previewed, and edited without leaving the conversation. It's live for all Gemini users on web starting today, with mobile support for PDFs.
Why it matters: This closes a major usability gap between Gemini and competitors like Claude and ChatGPT, and turns Gemini into a genuine productivity layer on top of Google Workspace rather than just a chat interface bolted on the side.

Data Centers Could Erase 15 Years of US Emissions Progress by 2030
A Sierra Club analysis finds that US power sector CO2 emissions could jump 42% by 2030 compared to 2025 levels — entirely driven by fossil fuel buildout to feed AI data center electricity demand. That's 618 million additional tons of CO2, undoing a decade and a half of hard-won reductions.
Why it matters: The AI industry's energy appetite is no longer an abstract future concern — it's on track to be one of the single largest drivers of US emissions growth this decade, and the political and regulatory blowback is coming.
Quick Hits
- →Deepgram acquired Daily, the company behind the open-source Pipecat voice AI framework, to beef up its real-time voice and video infrastructure. TechCrunch
- →Wabi, the AI agent that vibe-codes mini apps on demand, raised $55M at a $1.25B valuation and is rebranding itself as a "personal agent." TechCrunch
- →Microsoft disclosed for the first time that Azure hit $33.2B in quarterly revenue — and separately confirmed it booked $10B from its OpenAI commercial arrangements in a single quarter. TechCrunch
- →Meta quietly launched Style Sketch, an AI fashion app that scrapes your Instagram photos to build a virtual wardrobe and suggest outfits. TechCrunch
- →Agent identity startup Mintive — founded by Google Brain veterans — landed seed funding backed by Jeff Dean and Andrej Karpathy to help enterprises manage and authenticate AI agents. TechCrunch
AI TLDR