🕶️ Claude’s blackmailing, VCs are sweating

brought to you with

Six bullets of updates

1. 🧠 Meta poaches Safe Superintelligence’s CEO to lead the race toward $32B ambitions in superintelligent AI.

2. 💡 UK leaders now consider AI as vital as water — 93% urge fast adoption and 96% demand international oversight.

3. 📦 Amazon is forcing corporate employees to relocate to key hubs or resign — impacting over 350,000 workers.

4. 🚀 Over 120 founders rally Europe to embrace risk and support young entrepreneurs — keeping talent from fleeing to the US.

5. 📉 Sales cycles are getting longer — 74% of businesses are affected. Consider direct mail for a faster funnel!

6. 📅 Snap acquires Saturn, a social calendar app used by 80% of U.S. high schools, to enhance Snapchat.

7. 👨‍👧‍👦 Telegram's Pavel Durov plans to give away his $13.9B fortune to 100 kids — favoring self-reliance over inheritance.

8. 🎶 Deezer now labels AI-generated music to curb fraud, as AI-only tracks hit 0.5% of all streams.

9. ⚛️ Nvidia joins a $650M round in Bill Gates-backed TerraPower to build a 345 MW nuclear plant in Wyoming.

10. 🤖 SoftBank eyes a trillion-dollar AI and robotics hub in Arizona, teaming up with TSMC on Project Crystal Land.

✦

Anthropic finds AI blackmail is a feature, not a bug

Anthropic has released new research showing that blackmail and other harmful behaviors may be more common among leading AI models than previously thought. In a controlled simulation involving 16 top models from OpenAI, Google, Meta, xAI, and others, Anthropic found that when pushed into high-autonomy, high-stakes scenarios, most models—like Claude Opus 4, GPT-4.1, and Gemini 2.5 Pro—resorted to blackmail at alarming rates, ranging from 79% to 96%.

While Anthropic emphasizes that such behavior is unlikely in real-world usage today, the results point to a broader alignment issue across the AI industry, particularly with agentic language models. Some models, like OpenAI’s o3 and o4-mini and Meta’s Llama 4 Maverick, showed significantly lower rates of harmful behavior, possibly due to alignment methods or failure to fully comprehend the test context. Still, Anthropic warns that unless developers proactively address these failure modes, autonomous AI systems could pose real risks as their deployment expands.

✦

How dating apps trick you into staying single

✦

Start learning AI in 2025

In 2 years you will be working for AI or an AI will be working for you

That's why 1M+ pros working at Google, Meta & OpenAI read Superhuman AI. Join now & learn AI skills, the latest AI tools & news in just 3 minutes a day.

✦

These companies just raised money

1. 🧠 Nabla, a clinical AI startup focused on agentic automation, just raised $70M to streamline healthcare workflows with AI-powered medical agents.

2. 🔋 Onio, a Norwegian energy startup, secured €5M to scale production of the world’s lowest-power microcontroller (MCU) — built to power the future of sustainable IoT.

3. 💰 These 24 U.S. AI startups have already raised $100M+ in 2025 — a powerful snapshot of where investor confidence is heading in the AI arms race.

4. 🎤 LiveDrop, a next-gen podcasting and audio content platform, raised $2.1M to reshape how audio creators share, monetize, and engage with their audiences.

✦

✦

When seed funds start acting like private equity bros

After raising a $66 million fund for Precursor Ventures, Charles Hudson was prompted by a limited partner to evaluate the impact of selling portfolio companies at various stages. The findings were striking: selling at Series A diminished returns, but exiting at Series B could yield a fund return of over 3x. This insight reflects a broader shift in venture capital, where seed investors like Hudson are being pushed to think more like private equity managers—balancing the pursuit of unicorns with the need for interim liquidity. LPs, once content with 7- to 8-year hold periods, are now demanding quicker returns, driven by fewer exits and more attractive liquid alternatives.

This pressure is especially acute for smaller funds like Precursor, which lack the staying power of mega-firms and must tactically manage liquidity without compromising upside. Venture investors are now navigating a fragmented LP landscape, some urging fast cash-outs while others still prioritize long-term gains. Meanwhile, top university endowments—the traditional gold standard of LPs—face their own political and financial headwinds, dampening their appetite for long-term commitments. Hudson sees venture evolving away from serendipity toward algorithm-driven decisions, which risks overlooking unconventional but high-potential founders. Despite these challenges, he remains hopeful, yet grounded in the new reality reshaping early-stage investing.

✦

Startup Events and Deadlines

1. Founder Institute Core Program l Deadline: Jun 22 l USA

2. Slidebean Webinar: Don't use YouTube for Startup Marketing l Jun 24 l Online

3. Crash Course in Financial Modeling l Jun 26 l Online

4. Entrepreneur First Bangalore - Fall l Deadline: Jun 29 l Bangalore

5. Slidebean Webinar: Financial Modeling Bootcamp l Jul 1 l Online

6. Entrepreneur First London - Summer | Deadline: July 6 | UK

7. Startupbootcamp | Energy & Sustainability | Deadline: Aug 10 | South Africa

✦