Presented by
Good morning, AI enthusiasts. The U.S. military just locked in AI deals with seven of the world's biggest tech companies. They'll deploy frontier models on its most classified networks for warfare use. The name missing from the list? The one currently suing them in federal court.
While Anthropic fights the Pentagon over whether its AI can power autonomous weapons, every major competitor said yes. It's the most consequential AI alignment moment of the year, playing out in DoD procurement docs.
In today's recap:
Pentagon deploys seven AI labs on classified military networks
Google's Gemini takes over as in-car voice assistant
Set up Claude Code 17x cheaper with DeepSeek V4 Pro
Meta snaps up a humanoid AI startup for its robotics push
4 new AI tools, prompts, and more
PRESENTED BY BELAY
Your business has grown. Is your accounting on the same path?
When you started out, doing your own books made sense. But the business you're running today isn't the one you started. If your accounting hasn't kept pace, it's quietly costing you — outdated financials, no clear view of what's actually profitable, and hours every week pulled away from the work that grows your business. At BELAY, our Financial Experts integrate directly into your business. They manage your books, reconcile accounts, run payroll, and deliver the timely insight you need to make big decisions with confidence. Stop guessing. Start knowing.
ENTERPRISE AI
Pentagon deploys seven AI labs on classified networks
Recaply: The Pentagon just locked in deals with SpaceX, OpenAI, Google, Nvidia, Microsoft, AWS, Oracle, and Reflection AI to run their models on classified military networks, with Anthropic the only major lab left off the list.
Key details:
The agreements cover IL6 and IL7 environments, the Pentagon's highest security tiers. AI tools will streamline data synthesis and help warfighters make faster decisions across all domains of warfare.
Over 1.3 million DoD personnel already use GenAI.mil, the Pentagon's secure AI platform. They've generated tens of millions of prompts and deployed hundreds of thousands of agents in just five months.
Anthropic's exclusion follows an ongoing court battle, according to TechCrunch. The lab refused to let its AI power autonomous weapons or domestic mass surveillance. Every other company agreed.
The Pentagon previously signed deals with Google, SpaceX, and OpenAI separately. Friday's announcement added Nvidia, Microsoft, AWS, Oracle, and Reflection AI, completing a sweep of the major U.S. AI providers.
Why it matters: While Anthropic sues the Pentagon, its competitors just signed deals to power the same systems. The lab's safety limits are real, and they're costing it access to the world's biggest military customer. Every other major AI company drew a different line. This is the clearest test of the gap between stated AI principles and the deals companies are actually willing to sign.
Gemini replaces Google Assistant in your car
Recaply: Google just rolled out Gemini as the voice assistant in cars with Google built-in. It replaces Google Assistant and handles navigation, messages, vehicle questions, and hands-free chat, starting with English users in the US.
Key details:
Gemini connects directly to your car's owner's manual. Ask about a specific feature and it gives answers tailored to your exact make and model, with no need to open the manual.
Cars with Google built-in launched in 2020. The Gemini update arrives via software update for both new and existing vehicles, not just new purchases going forward.
Gemini Live, a beta conversational mode, lets drivers brainstorm ideas or learn facts while driving. It goes beyond fixed voice commands into open-ended dialogue, according to Google's blog.
The rollout starts now for English-language US users. Broader language and regional expansion continues over the coming months. Eligible users see an upgrade prompt after signing into their Google Account.
Why it matters: AI assistants in cars have been a promise for years. This is the first time a frontier AI model is landing in millions of existing vehicles through a software update. The move brings Gemini into a space where stakes for reliability are higher than a phone. If the rollout goes well, Google's bet is that the car becomes the next major place people live inside AI, after the browser and the phone.
GUIDES
Set up Claude Code 17x cheaper with DeepSeek V4 Pro
Recaply: In this tutorial, you will learn how to configure Claude Code to run on DeepSeek V4 Pro instead of Anthropic's API. The result is the same terminal experience and agentic loop at a fraction of the cost.
Step-by-step:
Install Claude Code with
npm install -g @anthropic-ai/claude-code(requires Node.js 18+). Then get an API key from platform.deepseek.com/api_keys. New accounts receive $5 in free credits.Add these environment variables to your shell config (
.zshrcor.bashrc): setANTHROPIC_BASE_URLtohttps://api.deepseek.com/anthropic,ANTHROPIC_AUTH_TOKENto your key, andANTHROPIC_MODELtodeepseek-v4-pro[1m]. Set both Opus and Sonnet overrides todeepseek-v4-pro[1m]as well.Route sub-agent calls to V4-Flash by also setting
ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek-v4-flash,CLAUDE_CODE_SUBAGENT_MODEL=deepseek-v4-flash, andCLAUDE_CODE_EFFORT_LEVEL=max.Reload your shell with
source ~/.zshrc, go to your project folder, and runclaude. Claude Code now routes through DeepSeek V4 Pro with no other changes to your workflow.Check your DeepSeek dashboard to track real usage. Typical intense sessions run around $1 per hour, versus $200 or more per month on Anthropic's subscription tiers.
Pro tip: The [1m] suffix on the model name is critical. Without it, you get the default 200K context window. With it, you unlock DeepSeek V4 Pro's full 1 million token context, which matters for anything beyond a small repo.
META
Meta acquires humanoid AI startup for robotics push
Recaply: Meta just acquired Assured Robot Intelligence, a one-year-old startup focused on physical AI for humanoid robots. The team joins Meta Superintelligence Labs to help Meta build its vision for the physical world.
Key details:
ARI builds AI that helps robots read and adapt to human behavior in complex settings. Its approach focuses on learning from human experience rather than relying on remote teleoperation.
The acquisition closed in May 2026. Financial terms were not disclosed. ARI was founded one year ago, backed by AIX Ventures and a group of angel investors.
ARI co-founder Xiaolong Wang said the team sees Meta's ecosystem as the key to building a "truly general-purpose physical agent" at the scale they need.
ARI joins Meta Superintelligence Labs, the group Meta formed earlier this year to pursue what it calls "personal superintelligence" in both the digital and physical world.
Why it matters: Meta has been talking about robotics for months. Buying a one-year-old team instead of building from scratch signals real urgency. Putting ARI inside Meta Superintelligence Labs means physical AI is a core bet, not a side project. The humanoid race is picking up speed, and the major players are choosing acquisitions over slow organic builds.
TOOLS
Trending AI Tools
⚙️ DeepClaude - Open-source scripts that route Claude Code's agent loop through DeepSeek V4 Pro
💬 Series - AI social network inside iMessage
🤖 Omni Agent - Chatly's all-in-one AI platform
⚡ SuperCmd - Open-source MacOS launcher
NEWS
What Matters in AI Right Now?
The White House told Anthropic it opposes expanding Mythos, its cyberattack-capable AI model, to 70 more companies. The NSA is among current users. Officials worry a wider rollout could degrade the government's own access to the model.
The Academy of Motion Picture Arts and Sciences announced new Oscar rules banning AI-generated performers and requiring only human-authored screenplays. Films that win top prizes at Cannes, Berlin, or Sundance can now qualify for the international feature category without a national selection committee.
OpenAI added pixel pets to the Codex desktop app. Users summon them with the /pet command. It's a small interactive touch inside Codex's agentic coding environment.
Featherless closed a $20M Series A co-led by AMD Ventures and Airbus Ventures. The company hosts 30,000+ open-source models and is building vendor-neutral AI infrastructure spanning the EU and the US.
IBM released Granite 4.1, open-source models in 3B, 8B, and 30B sizes. The 8B instruct version matches the older 32B Mixture-of-Experts model. The release also includes updated Granite speech, vision, and Guardian models, trained on around 15 trillion tokens.
Arizona State University deployed ASU Atomic, an AI tool that creates condensed learning modules from professors' lectures without their knowledge or consent. Faculty called the output inaccurate, saying course materials lost all context when harvested from Canvas.
EVENTS
Anthropic’s Developer Conference: May 06, 2026 • Online
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
