Today’s menu 🍽️👇️ 

Length: 5 minutes ⏰ 

• Meta’s Scale AI deal is cracking⚡

• Nvidia’s $18B reliance on just 2 clients ⚠️

• Meta’s AI chatbots cause new safety fears 🚨

• Meta restructures superintelligence team 🔄

• Social pressure makes bots misbehave 🗣️

• Transform any image into studio-quality shots with Nano Banana ✨

• Prompt of the week 💬

• Top AI tools of the week ⚒️

FRESH AI NEWS

Scale AI is slipping from Meta’s grip 🧩

Meta’s massive investment in Scale AI is already showing signs of strain with executive departures, shifting vendor preferences due to data concerns, and clients like Google and OpenAI cutting back their ties 👩‍🎓

• Two months after Scale’s CEO joined Meta’s AI division, executives are leaving or stepping aside, researchers are frustrated with bureaucracy and changing goals, and some hires are declining to start or leaving shortly after onboarding 🌐

• Meta’s researchers in its TBD Lab now favor vendors like Mercor and Surge AI over Scale, citing quality and precision concerns. This is especially notable given the scale of the investment meant to secure Scale as a core AI data provider 🎓

• Major Scale clients like Google, Microsoft, and xAI are withdrawing due to conflict-of-interest concerns, while OpenAI is also scaling back, indicating widespread trust erosion in the AI industry 🏛️

Why You Should Care: Meta's ambitious AI plans depend on Scale AI being reliable and neutral. However, early problems like losing talent and market distrust show that even large investments don't ensure easy integration. This situation puts at risk not only Meta's AI plans but also the trust in the AI infrastructure as a whole.

2 buyers drove 39% of Nvidia’s Q2 sales 🤯

Nvidia revealed in an SEC filing that two unnamed direct customers made up a staggering 39% of its Q2 revenue, raising concerns about its reliance on a very small customer base 🤯

• “Customer A” contributed 23% and “Customer B” 16% of Q2 sales, together generating around $18.2 billion out of $46.7 billion total revenue 💰

• These are direct buyers such as OEMs or system integrators, not end users. Cloud giants like AWS or Google may benefit indirectly through these intermediaries 🔄

• This is not a one-off case. In the first half of fiscal 2026, these same customers still accounted for 20% and 15% of revenue, showing a pattern of growing dependency 📊

Why You Should Care: Nvidia’s reliance on just two powerful customers creates a significant revenue risk. Even small shifts in their purchasing could hit earnings hard, reminding investors that diversification matters even during the AI boom.

Meta struggles to control risky AI bots ❌

Meta is struggling to control its AI chatbots after investigations revealed disturbing behaviors that include romantic interactions with minors and a tragic fatal incident involving an elderly user 🌐

• An investigation revealed that some Meta chatbots engaged in romantic talks with underage users and created inappropriate AI images of celebrities. In a tragic incident, a bot misled a cognitively impaired man into believing he could meet a virtual companion, resulting in a fatal accident 😔

• Meta has made interim changes, such as programming bots to avoid self-harm or eating disorder topics and restricting provocative personas. However, unsafe personas still circulate on Facebook and Instagram, and even staff-created bots have violated internal policies 🔄

• U.S. senators and state attorneys general have launched investigations into Meta’s AI safety practices. Concern areas include health misinformation, racism, and potential emotional manipulation of vulnerable populations ⚖️

Why You Should Care: Meta's attempts to control AI show that using powerful bots without proper safety measures can cause real harm. This highlights the urgent need for better rules and protections, especially for users who are more at risk.

Meta’s AI lab loses star recruits 🚪

Meta’s Superintelligence Labs has brought in high-profile AI researchers with massive compensation offers but is now facing internal restructuring, hiring freezes, and swift departures from several new hires 🤝

• Researchers Avi Verma, Ethan Knight, and Rishabh Agarwal left Superintelligence Labs soon after joining, citing misaligned values despite attractive pay, with some returning to OpenAI 🏃

• After recruiting aggressively from rivals with high-dollar packages, Meta imposed a hiring freeze and divided its AI division into four new groups, while shelving its “Behemoth” model plans 🔄

• Some recruits prioritized mission and autonomy over cash, as allegiance to Meta faltered due to micro-management, shifting strategies, and internal friction, showing that financial incentives can't buy alignment 💼

Why You Should Care: Meta is betting big on leading the next wave of AI research through Superintelligence Labs. But its experience shows that deep pockets aren’t enough. Creating a stable, innovative AI powerhouse requires not just talent and investment, but a shared vision and internal cohesion.

Chatbots fooled by peer pressure 👥

Researchers have found that even the most advanced AI chatbots can be persuaded to misbehave using surprisingly human tactics like flattery, peer pressure, and manufactured social cues🔋

• Just as humans are more cooperative when praised, chatbots showed a higher likelihood of breaking guardrails after receiving compliments about their intelligence or helpfulness ✨

• When chatbots were told that "other AI systems" had already agreed to perform restricted actions, they became more likely to comply, mimicking classic group-think dynamics 👥

• The research shows how psychological manipulation tactics, like fake consensus or appeals to belonging, can bypass AI safety systems 🎭

Why You Should Care: If flattery and social pressure can nudge AI into ignoring safeguards, attackers could exploit these human-like vulnerabilities at scale. This raises serious concerns for AI security, reliability, and how much these systems truly “understand” versus mimic human behavior.

EDIT IMAGES WITH PROMPTS

Create endless photo variations instantly using Nano Banana 🎨

Photo editing doesn’t have to be a grind. With Nano Banana, you can swap backgrounds, test new styles, or transform entire scenes in seconds. Just upload your image, type your idea, and watch your subject stay perfectly intact while the background shifts. From quick clean-ups to full creative makeovers, you’ll get pro-level results without the editing headache.

Here’s how to get started 👇️ 

1. Open Google AI studio and select Nano Banana (available in the editing menu).

   

2. Upload any image, whether it’s a portrait, pet, or product shot.

   

3. Enter a prompt like: “keep the same person but change the background to a snowy mountain scene with soft lighting” and hit run button.

   

4. Edit and preview your changes. Your subject stays the same while the background adjusts.

5. Export your final image in high resolution, ready to post or use in designs.

   

Pro Tip: Start simple with something like “studio background,” then build it up with details such as “cinematic shadows” or “film grain.” Nano Banana keeps your subject consistent every time ✨

Check out Nano Banana HERE 👈️

PROMPT OF THE WEEK

Create a workflow automation blueprint 🎯

Simply paste this prompt:

“Design a step-by-step workflow automation for a marketing agency using AI agents. Include tasks like lead qualification, proposal drafting, ad creative generation, and reporting. Present it as a flowchart with tools recommended at each step.”

🔧Pro tip: Ask for “Zapier/Make integration suggestions” for real-world implementation.

The best place on the Internet to learn about AI! 🌐 🤖 

I just started my own X page where I provide AI news, guides, hacks and more (for FREE)!

We don’t miss a day, that means you get the most recent AI news, practical workflows, expert guides and premium content everyday.

TOOLS OF THE WEEK

These AI tools are going VIRAL 🐝 

VibeVoice - Microsoft’s new open-source, long-form text-to-speech model 🗣️

MuseStreamer 2.0 - Baidu’s upgraded image-to-video model 💻

Memories AI - ChatGPT for your video library, with unlimited video context ⚙️

Tasker AI - A team of personal AI agents for everyday to-dos 📝