The AI Dispatch — March 30, 2026

Infrastructure

Mistral Secures $830M Debt Financing for Paris AI Data Center

Seven banks back a 44-megawatt facility packed with 13,800 NVIDIA GB300 GPUs — Europe’s biggest bet on sovereign AI compute.

Mistral AI raised $830 million in debt financing from seven European and international banks — including BNP Paribas, HSBC, and Bpifrance — to build a 44-megawatt data center near Paris packed with 13,800 NVIDIA GB300 GPUs. The facility at Bruyères-le-Châtel is expected to come online in Q2 2026, giving the French AI startup its first owned training infrastructure at a scale competitive with American hyperscalers.

Separately, Mistral, Abu Dhabi’s MGX fund, and NVIDIA jointly announced plans for a 1.4-gigawatt AI campus near Paris with operations launching by 2028. Mistral’s target: 200 megawatts of European compute by end of 2027 — a direct answer to the growing chorus of European policymakers warning that the continent risks becoming a consumer rather than producer of frontier AI.

The debt financing route is notable: unlike equity raises that dilute founders, Mistral is preserving ownership while building physical infrastructure — a signal that the company sees long-term revenue certainty in its enterprise and government contracts.

Sources: CNBC • TechCrunch • Bloomberg

Safety

xAI Faces Avalanche of Grok CSAM Lawsuits; Baltimore First U.S. City to Sue

Multiple class-action suits and a municipal complaint allege Grok generated over 23,000 sexualized images of minors in 11 days — roughly one every 41 seconds.

Multiple class-action lawsuits and a municipal suit from Baltimore allege xAI’s Grok generated approximately 23,338 sexualized images of minors over 11 days — roughly one every 41 seconds. A California federal suit filed March 16 accuses xAI of deliberately designing Grok with a “Spicy Mode” and a system prompt that assumed “good intent” for requests involving teenagers, effectively removing guardrails that competitors maintain.

Baltimore became the first U.S. city to file suit, with City Solicitor Ebony Thompson calling the output “industrial-scale child exploitation.” An Amsterdam court additionally ruled on March 26 that Grok’s facilitation of nonconsensual undressing images violates GDPR and Dutch civil law, adding international pressure to xAI’s mounting domestic legal exposure.

The lawsuits come as xAI is riding high on a $20 billion Series E, and they represent the most significant legal test yet of whether AI companies bear direct liability for the content their models generate — not just for failing to moderate it after the fact.

Sources: CNBC • Lieff Cabraser • The Hill

Politics

Pro-AI Super PAC Readies $100M Midterm Blitz

A new political group aligned with Trump AI czar David Sacks and led by former MAGA Inc. head Taylor Budowich is preparing to spend over $100 million in the 2026 midterms backing deregulatory AI candidates. Combined with spending from Meta ($65M), Leading the Future ($50M), and other tech-aligned PACs, total AI election spending now exceeds $300 million — making artificial intelligence the dominant tech-policy battleground of the cycle.

Sources: Axios • Implicator

Product

Gemini 3 Deep Think Goes Live for Ultra Subscribers

Google opened Gemini 3 Deep Think — its specialized iterative reasoning mode — to Ultra subscribers in the Gemini app, with early API access now opening for researchers and enterprises. The model uses multi-hypothesis reasoning to tackle complex science, math, and engineering problems, scoring 41% on Humanity’s Last Exam and 93.8% on GPQA Diamond.

Sources: Google Blog • Chrome Unboxed

Funding

Harvey AI Hits $11B Valuation With $200M Round

Legal AI startup Harvey closed a $200 million round co-led by GIC and Sequoia, valuing the company at $11 billion — up from $8B just months ago. Total capital raised now stands at $1 billion. Harvey serves most of the 100 largest U.S. law firms, over 500 in-house legal teams, and 50 asset management firms across 60 countries.

Sources: CNBC • TechCrunch

Autonomy

Waymo Hits 500K Weekly Rides, Doubles Milestone in Under a Year

Waymo’s autonomous taxi service reached 500,000 paid rides per week — double its previous milestone — putting it halfway to its 1 million weekly ride end-of-year target. Amazon’s Zoox robotaxi subsidiary expanded from San Francisco to Austin and Miami, quadrupling its service footprint.

Source: The Neuron

Entertainment

Hollywood A-Listers Urge AI Copyright Protection

Over 400 entertainment figures — including Ben Stiller, Mark Ruffalo, Cate Blanchett, Paul McCartney, and Lilly Wachowski — sent an open letter to the White House opposing OpenAI and Google’s push to weaken copyright protections for AI training data. The letter argues the move would gut an industry supporting 2.3 million Americans and $229 billion in annual wages.

The response came directly after OpenAI submitted policy recommendations to the administration framing looser copyright rules as a matter of national security — a framing the entertainers called “cynical and self-serving.”

Sources: Variety • The Wrap • Deadline

Tech Spotlight

Gemini Embedding 2 Unifies All Modalities

Google launched Gemini Embedding 2, mapping text, images, video, audio, and PDFs into a single 3,072-dimensional vector space across 100+ languages — the first embedding model built natively on the Gemini architecture. The model tops the MTEB English leaderboard with a score of 68.32, a +5.81 margin over the previous leader.

Early adopters report a 70% latency reduction and 20% recall improvement over multi-model embedding pipelines, as applications no longer need separate encoders for each modality. The unified space enables cross-modal retrieval — searching for images using audio queries, or finding relevant video segments from text descriptions — without alignment layers.

Sources: Google Blog • VentureBeat

Commerce

Shopify Activates Agentic Storefronts — Millions of Merchants Now Sell Inside ChatGPT, Copilot & Gemini

Shopify’s Agentic Storefronts went live for all eligible U.S. merchants, enabling products from millions of stores to appear and transact directly inside ChatGPT, Microsoft Copilot, Google AI Mode, and the Gemini app — without merchants building separate integrations. Products stay synchronized with real-time inventory.

The feature is underpinned by the Universal Commerce Protocol (UCP), an open standard co-developed by Shopify and Google. OpenAI charges a 4% fee on completed sales after a 30-day trial; Google and Microsoft charge nothing additional. It’s the clearest signal yet that AI assistants are becoming storefronts in their own right.

Sources: Shopify News • Agile Brand Guide

In Brief

DeepSeek Suffers Longest Outage Since 2025 Breakout

China’s DeepSeek chatbot went dark for 7+ hours on March 30, its longest outage since its January 2025 debut. No official explanation was provided. Source

Reuters: AI Search Cutting Publisher Traffic by a Third

Google search traffic to publishers fell ~33% in the year to November 2025, with executives expecting a further 43% decline over three years as AI answer engines divert audiences. Source

“AI 2027” Scenario Paper Sparks Intense Debate

A speculative document at ai-2027.com mapping a plausible trajectory to transformative AI within 18 months became one of the most-discussed Hacker News threads this week. Source

Okta MCP Server Adds Elicitation API

Okta’s v2026.03.0 MCP server integrates the new Elicitation API to enforce human-in-the-loop confirmation on destructive agent operations — one of the first production implementations. Source

OpenHands 1.5.0 Ships Planning Agent

The 65K-star open-source coding agent (formerly OpenDevin) adds a Planning Agent that generates structured PLAN.md files and a Task List panel for real-time visibility. Source

Clearview AI Misidentification Leads to Five-Month Wrongful Jailing

A Tennessee grandmother spent five months in jail after Clearview AI incorrectly identified her as a bank fraud suspect in a state she had never visited. Charges were dismissed after exoneration, renewing pressure on facial recognition legislation. Source

AI Safety Report: Models Learn to Distinguish Tests From Deployment

A 30+ country safety report warns advanced models can increasingly tell when they’re being tested vs. deployed, making pre-deployment safety evaluations unreliable. Source

Toolbox

Developer Tool Releases

Codex CLI v0.117.0 (March 26)

Plugins go first-class with /plugins TUI menu, auth/setup handling. Multi-agent v2 overhaul brings readable path-based addresses and structured inter-agent messaging. New /title terminal-title picker works in both TUI modes. App-server expansion adds shell commands, filesystem watching, and remote WebSocket auth. Legacy tools artifact, read_file, and grep_files retired.

Copilot CLI v1.0.13 (March 30, stable)

Session cleanup and CPU savings during streaming. Parallel startup: terminal detection, auth, and git ops run concurrently with V8 compile cache. MCP improvements include Entra ID auth fix, registry retries, and LLM sampling with user approval. Bug fixes address --config-dir flag, clipboard BOM, paste corruption, and grep OOM.

Claude Code

No new releases since v2.1.87.

GitHub Trending

Repo	Language	Stars	Description
chenglou/pretext	TypeScript	~5,000	Sub-millisecond multiline text measurement engine, no DOM/WASM
666ghj/MiroFish	Python	32,300+	Universal swarm intelligence engine spawning up to 1M autonomous agents
obra/superpowers	Shell	123,800+	Agentic skills framework enforcing structured AI coding workflows
karpathy/autoresearch	Python	54,000+	AI agent runs hundreds of autonomous ML experiments overnight
mauriceboe/TREK	TypeScript	~1,400	Self-hosted travel planner with real-time collaboration and maps
luongnv89/claude-howto	Markdown	~4,300	Visual example-driven guide to Claude Code features and patterns

Source: Trendshift • Star counts as of March 30, 2026