Tech Digest – March 10, 2026
AI Governance & Geopolitics
Anthropic Sues the Pentagon; White House Reportedly Preparing Executive Order to Strip Its AI from Federal Operations
Anthropic filed suit to block the Pentagon from placing it on a national security blacklist, arguing the designation violates its free speech and due process rights. The dispute stems from Anthropic’s refusal to grant blanket permission for its models in autonomous weapons and mass surveillance. In response, the White House is reportedly preparing an executive order to remove Anthropic’s AI from federal operations entirely. OpenAI announced a new Pentagon deal within hours.
Note: This isn’t a procurement footnote. It’s the first major legal test of whether an AI company can set limits on how a government uses its technology — and what happens when it tries.
Institutional Productivity & Platform Shifts
Microsoft Ships Copilot Cowork, Google Embeds Gemini Across Workspace — The Two Dominant Office Suites Are Now AI-Native
Microsoft launched Copilot Cowork, integrating Anthropic’s Claude Cowork into Microsoft 365 alongside its own Copilot. The move gives enterprise users agent-level AI inside the same productivity suite most institutions already run. Separately, Google rolled out Gemini across Workspace with a “Help me create” drafting tool. Meanwhile, GPT-5.4 Thinking set a new state-of-the-art on LiveBench at 80.28%, showing the capability curve behind these integrations is still steepening.
Note: If your institution uses Microsoft 365 or Google Workspace, AI just became a default feature of your existing licence. The question is no longer whether staff will encounter these tools — it’s whether you have a policy for when they do.
Sources: Microsoft, Google, LiveBench
One Non-Technical Marketer Ran Anthropic’s Entire Growth Operation for 10 Months — and AI Agents Now Review Code for $15–25 per Pull Request
Austin Lau, a growth marketer at Anthropic who had never opened a terminal, ran paid search, paid social, app stores, email marketing, and SEO using Claude Code for 10 months. He built a Figma plugin for ad variations and a Google Ads workflow — all without writing code. Separately, Anthropic launched Code Review, a service that dispatches multiple agents per pull request to catch bugs, averaging $15–25 per review with per-token billing. Internal testing showed developers rejected fewer than 1% of issues found.
Note: A single non-technical employee covering six marketing functions isn’t a productivity tip — it’s a staffing model. For anyone building a team or writing a headcount plan, the arithmetic has changed.
Sources: Anthropic, Anthropic Code Review
Capital & Infrastructure
Amazon Targets $37–42B Bond Sale for AI Data Centres; Murati’s Chip Deal and LeCun’s AMIL Signal the Scale of What’s Being Built
Amazon kicked off an 11-part bond offering targeting $37–42 billion to fund AI data centre expansion — likely one of the largest corporate bond sales in history. Mira Murati’s Thinking Machines Lab signed a chip supply deal with Nvidia worth tens of billions of dollars, planning to deploy over 1 GW of next-generation Vera Rubin chips. In Europe, Yann LeCun’s world-model startup AMIL raised a $1.03 billion seed round at a $3.5 billion pre-money valuation — the largest seed round the continent has ever seen.
Note: These are infrastructure commitments, not funding announcements. A $42 billion bond sale to build data centres is a 15-year bet on demand. Any institution planning digital infrastructure on a 5-year horizon is operating in a market shaped by these decisions.
Sources: Bloomberg, Financial Times (Murati), Financial Times (AMIL)
Enterprise AI Stack
OpenAI Acquires Promptfoo for AI Vulnerability Detection; Nvidia Pitches NemoClaw as Open-Source Agent Platform
OpenAI acquired Promptfoo, a tool for detecting vulnerabilities in AI systems — adding security testing to its enterprise stack. Nvidia announced NemoClaw, an open-source platform for building and deploying enterprise AI agents. Both moves signal that the AI toolchain is maturing past experimentation: the focus is now governance, security, and scalable deployment.
Note: When the two largest AI infrastructure providers both invest in agent governance in the same week, the procurement question shifts. It’s no longer “should we use AI agents?” — it’s “which agent framework, with what guardrails, and who audits them?”
Workforce Signals
UK Startups Now Average 2.7 Jobs Each as AI and Freelancers Replace Permanent Headcount
Bloomberg reports that UK startups now generate an average of just 2.7 jobs each, as founders increasingly substitute AI tools and freelance contracts for permanent employees. The trend compounds a broader shift: new companies are launching leaner than ever, and the jobs they do create look different from a decade ago.
Note: Employment programs and economic development strategies built on the assumption that startups create 10–15 jobs each need updating. The multiplier has changed.
Sources: Bloomberg
Robotics & Physical AI
TUM and NEURA Robotics Plan World’s Largest Robotics Training Centre at Munich Airport
The Technical University of Munich and NEURA Robotics announced plans for a 2,300 m² robotics training centre at Munich Airport — the world’s largest. The facility will combine academic research with industrial deployment, training the workforce that will operate alongside AI-powered robots in logistics, manufacturing, and public services.
Note: Europe’s largest robotics training investment is happening in Germany, not in a research paper. This is the physical AI equivalent of a hyperscaler data centre announcement — workforce infrastructure for what comes next.
Sources: TUM
Foundation Models Move Heavy Machinery; Amazon Targets 500 Million Drone Deliveries; FAA Greenlights Eight eVTOL Proposals
Actor Labs successfully ran the π0.5 foundation model on an excavator after labelling just 200 trajectories — demonstrating that general-purpose AI models can now operate heavy equipment with minimal training data. Amazon confirmed its target of 500 million drone deliveries per year by decade’s end, with expansion underway across multiple US cities. The FAA selected eight eVTOL proposals spanning air taxis, cargo, emergency response, and autonomous flight, signalling a regulatory green light for commercial urban air mobility.
Note: Foundation models controlling excavators with 200 training examples is a procurement signal, not a demo. The threshold for automating physical tasks just dropped by orders of magnitude.
Sources: Actor Labs (Lane Burgett), Amazon, US DOT
Defence & Security
Directed Energy Warfare Goes From Theory to Deployment: Havana Syndrome Devices Confirmed, Ukraine Fields Laser Counter-Drone Systems
CBS News reported that the US military tested directed energy devices linked to “Havana Syndrome” on rats and sheep, confirming the technology exists and produces the reported neurological effects. On the battlefield, Ukrainian forces are now using laser systems to destroy the fibre optics on Russian reconnaissance drones — a practical, low-cost counter-drone measure that marks the mainstreaming of directed energy in active conflict.
Sources: CBS News
