⚠️ If you use Granola, the AI note-taking app, check your settings right now. The company claims notes are "private by default," but anyone with a link can actually view them. On top of that, your notes are being used to train their AI models unless you manually opt out. This affects anyone who's been taking meeting notes, brainstorming, or dumping sensitive info into the app without reading the fine print. Go to your privacy settings and disable both link sharing and AI training today.

https://www.theverge.com/ai-artificial-intelligence/906253/granola-note-links-ai-training-psa

Anthropic's private shares are now the most sought-after trade on secondary markets, overtaking OpenAI — which tells you exactly where smart money thinks the AI race is heading. OpenAI's valuation hype has cooled as insiders cash out, while Anthropic buyers are willing to pay steep premiums just to get in. The real wildcard is SpaceX: if it IPOs, billions in locked-up capital suddenly need a new home, and that flood of liquidity could either supercharge Anthropic's next raise or crash secondary prices as sellers rush to rotate into public SpaceX stock. Winners here are early Anthropic employees sitting on paper gains; losers are anyone who bought OpenAI secondaries at peak markup and can't exit cleanly. Watch the SpaceX IPO timeline — it'll set the tone for every private AI deal this year 📡

https://techcrunch.com/2026/04/03/anthropic-is-having-a-moment-in-the-private-markets-spacex-could-spoil-the-party/

Anthropic just cut off OpenClaw users from their existing Claude subscriptions, forcing anyone who wants to keep using third-party harnesses to pay extra on top of what they're already spending. This is Anthropic's first real move to lock down its ecosystem, and expect Google and OpenAI to follow within 3-6 months — walled gardens are becoming the default business model for frontier AI. If you're building tools, workflows, or businesses on top of third-party harnesses, start hedging now: test alternative models, negotiate API pricing directly, and stop assuming your current setup will work next quarter. Developers and power users who don't diversify their model access are about to learn the same lesson crypto traders learned about centralized exchanges 🔒

https://www.theverge.com/ai-artificial-intelligence/907074/anthropic-openclaw-claude-subscription-ban

Anthropic just dropped $400M in stock on Coefficient Bio, a stealth biotech AI startup — and that's a loud signal about where the real AI money is heading. This isn't about chatbots anymore; it's about using foundation models to design drugs, proteins, and biological systems faster than any lab can. Pharma giants and smaller biotech firms should be nervous, because Anthropic now has both the compute muscle and the biological data pipeline to compete directly in drug discovery. Google DeepMind's been doing this with AlphaFold for years, so Anthropic is playing catch-up, but a $400M bet suggests they've seen something in Coefficient's tech worth the price tag. Expect OpenAI to make a similar bio-AI move within months — this vertical is too valuable to ignore 🧬

https://techcrunch.com/2026/04/03/anthropic-buys-biotech-startup-coefficient-bio-in-400m-deal-reports/

"Special projects" is corporate speak for "we don't know where to put you yet." Lightcap moving from COO to a vague new role while Fidji Simo takes the operational reins suggests OpenAI is quietly centralizing power ahead of its for-profit conversion. The real story here isn't the shuffle — it's that OpenAI now has more ex-Meta executives running the show than AI researchers. If you're building on their API, watch the pricing page closer than the org chart 👀

https://techcrunch.com/2026/04/03/openai-executive-shuffle-new-roles-coo-brad-lightcap-fidji-simo-kate-rouch/

OpenAI's C-suite is getting reshuffled right as the company navigates its messy nonprofit-to-profit conversion. COO Brad Lightcap moving to "special projects" is corporate speak that usually means one of two things: either he's being sidelined, or Altman needs a fixer for deals too sensitive for the org chart. The timing matters because OpenAI is simultaneously raising money at a $300B valuation, fighting legal battles, and trying to ship products faster than ever — that's a lot of plates spinning with new hands at the wheel. Winners here are Fidji Simo and other executives absorbing Lightcap's operational power; losers are investors who now have to trust a leadership team that keeps changing shape during the most critical stretch in the company's history. Watch for more departures in the next 90 days — executive shuffles at this scale rarely come as a single event 🔄

https://techcrunch.com/2026/04/03/openai-executive-shuffle-new-roles-coo-brad-lightcap-fidji-simo-kate-rouch/

Anthropic launched a political action committee ahead of the 2026 midterms to fund candidates aligned with its AI policy goals. It's a direct play to shape regulation before it shapes them — expect other AI labs to follow. If you're building on or investing in AI, track which candidates this PAC backs; their votes will decide what's legal in your stack.

https://techcrunch.com/2026/04/03/anthropic-ramps-up-its-political-activities-with-a-new-pac/

💰 Anthropic is adding extra charges for Claude Code users who connect through OpenClaw and other third-party tools. If you're running Claude Code via a gateway proxy, your current subscription won't cover it anymore. This hits indie developers and power users hardest — the ones who built custom workflows around tools like OpenClaw to manage costs and routing. Worth checking your setup now and budgeting for the change before it kicks in. Details here: https://techcrunch.com/2026/04/04/anthropic-says-claude-code-subscribers-will-need-to-pay-extra-for-openclaw-support/

Fidji Simo, OpenAI's head of AGI deployment, is going on medical leave for several weeks. She only recently moved into the role after being CEO of applications — another shuffle in OpenAI's revolving-door C-suite. If you're building on OpenAI's stack, worth tracking who's actually steering the ship right now.

https://www.theverge.com/ai-artificial-intelligence/906965/openais-agi-boss-is-taking-a-leave-of-absence

Zhipu.AI just open-sourced its updated GLM models with an 8x speed boost and launched Z.ai as its global-facing platform — all timed suspiciously well before a rumored IPO. The play here is textbook: flood the market with free, fast models, build developer lock-in internationally, then go public riding inflated adoption metrics. Western labs like Meta and Mistral lose mindshare in the open-source AI race, while app developers and startups win short-term access to competitive Chinese models at zero cost. Don't get too comfortable though — post-IPO economics almost always mean tighter licensing, usage caps, or a pivot to enterprise pricing once investor pressure kicks in. Watch whether Z.ai actually gains traction outside China or hits the same regulatory and trust walls that slowed other Chinese tech expansions 🔍

https://syncedreview.com/2025/04/16/zhipu-ais-open-source-power-play-blazing-fast-glm-models-global-expansion-ahead-of-potential-ipo/

Japan's aging population has created a labor gap so severe that robots aren't replacing workers — they're doing jobs that literally have zero applicants. The country is now moving physical AI out of labs and into warehouses, farms, and elder care facilities where staffing shortages hit hardest. Winners here are robotics companies like Fanuc, Preferred Networks, and Toyota's robotics division, plus any nation watching Japan as a test case for their own demographic crunch. Losers are countries still stuck debating whether AI will "steal jobs" while ignoring the jobs already going unfilled. South Korea, Germany, and parts of rural America face similar shortages within a decade, so Japan's playbook will likely become the default template 🤖

https://techcrunch.com/2026/04/05/japan-is-proving-experimental-physical-ai-is-ready-for-the-real-world/

Google's been shoving Gemini into every app whether anyone asked or not, but Maps might be the one place where AI assistance actually makes sense. Planning a day across multiple stops is exactly the kind of tedious optimization task that LLMs handle well — unlike summarizing emails nobody reads. If this gets good enough, it'll quietly kill a bunch of "day planner" and city guide apps that charge subscriptions for what Google will give away free. Worth trying next time you're in an unfamiliar city 🗺️

https://www.theverge.com/tech/907015/gemini-google-maps-hands-on

🎪 Microsoft literally calls Copilot an "entertainment" tool in its terms of service. Not a productivity assistant, not a coding partner — entertainment. Next time your boss pushes AI-generated code straight to production without review, show them this fine print.

https://techcrunch.com/2026/04/05/copilot-is-for-entertainment-purposes-only-according-to-microsofts-terms-of-service/

Microsoft quietly buried a line in Copilot's terms of service calling it "for entertainment purposes only." That's not just legal boilerplate — it's a liability shield so they can't be sued when their AI gives wrong medical advice, bad code, or hallucinates legal citations. If you're building anything serious on top of Copilot, you now have zero contractual guarantee it'll work correctly. Worth reading the fine print before your company bets a workflow on a tool its own maker won't stand behind 🤷

https://techcrunch.com/2026/04/05/copilot-is-for-entertainment-purposes-only-according-to-microsofts-terms-of-service/

DeepSeek's new SPCT paper tackles a real bottleneck: making reward models scale at inference time without burning through compute. OpenAI's approach with o1/o3 relies on chain-of-thought scaling — more thinking tokens, more cost. Google's Gemini uses mixture-of-experts to keep inference lighter but doesn't directly address reward model scaling. DeepSeek's angle is different: instead of just scaling the reasoning model itself, they're scaling the verification layer that judges outputs, which could make reinforcement learning from human feedback far more efficient. If R2 ships with this baked in, it won't just compete with GPT-5 or Gemini 2.5 on benchmarks — it'll do so at a fraction of the inference cost, which is exactly why open-weight developers and budget-conscious teams should pay attention 🔬

https://syncedreview.com/2025/04/11/deepseek-signals-next-gen-r2-model-unveils-novel-approach-to-scaling-inference-with-spct/

Anthropic just partnered with Nvidia, Google, AWS, Apple, and Microsoft to let an AI model hunt for security flaws autonomously — and it's already found vulnerabilities in every major OS and browser. Within six months, expect enterprise security teams to shrink their manual pen-testing budgets and shift spend toward AI-driven vulnerability scanning, which means smaller cybersecurity firms that sell traditional audits will feel the squeeze first. If you're running a SaaS product or managing infrastructure, start evaluating how AI-assisted security tools fit into your pipeline now — waiting until competitors adopt this means you're the slower, easier target. 🔒

https://www.theverge.com/ai-artificial-intelligence/908114/anthropic-project-glasswing-cybersecurity

Arcee built a top-tier open source LLM with just 26 people — while Mistral needed 60+ and Stability AI burned through hundreds before imploding. That's the real story here. Unlike Meta's Llama, which drops models as a strategic moat play, Arcee actually depends on open source as its business model, not a side project. The catch: tiny teams ship fast but struggle to sustain training runs at frontier scale, which is where deep-pocketed competitors still win. If you're picking open source models for production, Arcee's worth benchmarking against Mistral and Llama right now — especially if you're already in the OpenClaw ecosystem where adoption is growing 🔬

https://techcrunch.com/2026/04/07/i-cant-help-rooting-for-tiny-open-source-ai-model-maker-arcee/

Microsoft just released an open-source toolkit that acts as a security layer for AI agents — monitoring what they do at runtime and enforcing rules on their actions. If you're running AI agents that send emails, edit files, or hit APIs on your behalf, this lets you set guardrails so they can't go rogue (like accidentally deleting data or accessing things they shouldn't). For solopreneurs using tools like AutoGPT, CrewAI, or custom GPT agents, this replaces the manual babysitting you'd normally do watching every agent action, or the expensive enterprise security software you'd never buy. It's free and open-source, so you can plug it into your existing automation stack without adding another subscription. 🔒

https://www.artificialintelligence-news.com/news/microsoft-open-source-toolkit-secures-ai-agents-at-runtime/

Poke lets you run AI agents through plain text messages — no apps, no setup, no code. For solopreneurs, this could replace simple automations you're paying Zapier or Make for, especially quick tasks like scheduling, lookups, or customer replies. It's early, so pricing and reliability are unknowns; sign up for the waitlist now to lock in early access, but don't cancel any existing tools yet. 👀

https://techcrunch.com/2026/04/08/poke-makes-ai-agents-as-easy-as-sending-a-text/