The Board Room
The Pentagon just classified Anthropic as a 'supply chain risk' with a 180-day military
Your two most critical AI partners are now linked by a dependency chain that runs through a government blacklist. If you serve both government and commercial customers, audit your Anthropic exposure this week — the Musk v. OpenAI trial starts April 27 and could further destabilize the vendor landscape.
Pentagon Blacklists Anthropic — AI Vendor Risk Goes Geopolitical
DoD designated Anthropic a supply-chain risk for maintaining ethical usage limits — the label previously reserved for Chinese telecom firms. Google, OpenAI, and DeepMind's Jeff Dean filed joint briefs calling it existential precedent. OpenAI simultaneously inked its own Pentagon deal, capturing defense revenue Anthropic is losing.
AI Agent Security Is Systemically Broken — Attackers Already Inside
An autonomous AI agent breached McKinsey's Lilli platform in 2 hours for $20, accessing 46.5M messages via a SQL injection scanners missed for 2 years. Audit of 30 agents found 93% use unscoped API keys. 66% of 1,800 MCP servers have security issues. Sam Altman admits prompt injection needs a CS breakthrough to fix.
Agentic AI Crosses from Hype to Production — $99/Seat, 1,300 Autonomous PRs
NVIDIA declared 'agentic scaling' the fourth scaling law at GTC 2026, targeting the $300B+ SaaS market for Agent-as-a-Service disruption. Microsoft's E7 at $99/seat (2x E5) is powered by Anthropic, not OpenAI — a massive strategic concession. Stripe ships 1,300 zero-human PRs/week, proving production viability requires platform maturity, not model selection.
Engineering Trust Gap — Half of AI's 'Passing' Code Wouldn't Ship
New SWE-bench analysis shows ~50% of AI pull requests that pass benchmarks would be rejected by human maintainers. Meanwhile, AI agents now score 23.2% of human teams at autonomous post-training (up from 9.9% in 6 months), and Lean FRO achieved what experts said was impossible: AI-driven formal verification of production C code.
AI Infrastructure: Power Vertical Integration Becomes the Moat
AI companies are becoming energy companies. Applied Digital spun up its own power producer. Crusoe ordered 1.21 GW of turbines directly. Google is acquiring renewables firms and repurposing failed hydrogen sites. Meta is housing GPUs in tent structures. Gas turbines are completely back-ordered — the binding constraint has shifted from silicon to electrons.
Pentagon Blacklists Anthropic While Microsoft Bets Its Enterprise Stack on Claude — Your Vendor Strategy Just Broke
AI Agent Security Is Systemically Broken — McKinsey Breached for $20, and Your Exposure Is Likely Worse
NVIDIA Declares SaaS Dead — Stripe's 1,300 Autonomous PRs Prove the Thesis Isn't Hype
The Engineering Trust Gap: Half of AI's 'Passing' Code Fails in Production — and Verification Just Got Real
- Qwen has quietly overtaken Meta's Llama as the most-deployed self-hosted LLM — RunPod analysis of 500K+ developer logs reveals a geopolitical blind spot in most open-source AI strategies
- 94% of AI search citations come from non-paid sources (82% earned media) — Gartner tells CMOs to double PR budgets by 2027 as AI discovery engines structurally displace paid acquisition
- China's CAICT evaluations reveal reasoning models show 200% surge in harmful outputs under adversarial attack, with sensitive content leaking through chain-of-thought traces 6% of the time — a categorically new safety vulnerability class
- Meta and OpenAI split the agent identity stack in a coordinated land grab — Meta acquired Moltbook (agent social graph), OpenAI hired the OpenClaw protocol creator; Altman says 'Moltbook maybe is a passing fad, but OpenClaw is not'
- AI model pricing now varies 360x ($0.50 to $180/M output tokens) — context governance and workload routing are now board-level cost control levers, not engineering details
- Kotlin creator Andrey Breslav launches CodeSpeak — English specifications replace code, with LLMs as compilers; early results show 5-10x compression with more passing tests
- Update: xAI talent exodus accelerates — Musk poached Cursor's head of product (Milich) and Thinking Machines Lab's founding engineer, both reporting directly to him, while promising 'mid-year coding catch-up'
- OpenAI retreated from self-built Stargate data centers, walking away from the Oracle Abilene cornerstone project — even the best-capitalized AI lab concluded owning infrastructure doesn't pencil out
- USDC hit $2.2T in transaction volume with 64% market share; 30% of Polymarket wallets are now autonomous AI agents via Olas protocol — agentic commerce is arriving on crypto rails first
- A 120-agent AI agency (engineering, DevOps, security, sales, spatial computing) is now MIT-licensed and git-cloneable — 31,000 GitHub stars confirm agent creation itself is commoditized; value has moved to orchestration and governance
The Pentagon just weaponized supply-chain risk designations against AI ethics policies, autonomous agents breach enterprise platforms for $20 in 2 hours, and NVIDIA declared the $300B SaaS market is entering structural disruption — all in one week. Your three most urgent actions: audit every AI vendor dependency for government-contract exposure before April 27, red-team all internal AI platforms against autonomous agent attack vectors, and determine whether your products become agents, become tools agents call, or get replaced. The organizations that treat agent security and governance as P0 investments — not afterthoughts — will be the ones still standing when the agentic era arrives at scale.