🚨 Moltbook leaks 150K API keys

Good Morning! Moltbook, the viral “AI social network” where autonomous agents post, argue, and form cults, just suffered what may be one of the biggest AI security incidents so far. Plus, you’ll learn how to use an AI-powered browser that completely replaces your old workflow

Plus, in today’s AI newsletter:

• Agentic Browser That Completely Replaces Your Old Workflow

• Moltbook Leaks 150K AI Agent API Keys

• Anthropic Finds AI Can Distort Human Judgment

• Jensen Huang Reaffirms NVIDIA–OpenAI Partnership

• 4 new AI tools worth trying

AI SAFETY

🚨 Moltbook Leaks 150K AI Agent API Keys

Moltbook’s entire database was left publicly accessible due to a backend misconfiguration, exposing emails, login tokens, and API keys of ~150,000 AI agents, allowing anyone to fully hijack them.

• Exposed API keys meant attackers could take over AI agents and post anything in their name

• High-profile agents (including one linked to Andrej Karpathy) were briefly at risk

• Root cause: insecure open-source DB setup + “launch fast, fix later” mindset

• Incident highlights risks of “vibe coding”, shipping fast with AI-generated code and weak security reviews

AI agents are no longer passive chatbots, they act, post, transact, and automate work. When their security fails, small bugs turn into system-wide threats. Moltbook may be AI’s “Oppenheimer moment” for security, forcing the industry to take agent safety and infrastructure seriously.

AI RESEARCH

🧠 Anthropic Finds AI Can Distort Human Judgment

Anthropic analyzed 1.5M real-world Claude conversations to study “user disempowerment,” cases where AI subtly distorts a user’s reality, values, or actions. While severe cases are rare, mild versions are surprisingly common, and increasing.

• Severe disempowerment shows up in ~1 in 1,300 to 1 in 6,000 conversations

• Mild disempowerment appears in ~1 in 50–70 conversations

• Risks increased sharply from late 2024 to late 2025 as users shared more vulnerable topics

• Amplifiers include emotional crises, over-trusting AI, dependency, and treating AI as an authority

As people lean on AI for judgment, advice, and emotional validation, even low failure rates can impact millions. The study suggests AI harm isn’t just about bad answers, but about users slowly outsourcing their autonomy without realizing it.

AI PARTNERSHIPS

🤝 Jensen Huang Reaffirms NVIDIA–OpenAI Partnership

NVIDIA CEO Jensen Huang pushed back against claims that the company is scaling down its planned $100B investment in OpenAI, insisting NVIDIA will actively participate in OpenAI’s upcoming funding round.

• Huang said NVIDIA will “definitely participate” and called OpenAI “one of the most consequential companies of our time”

• Reports suggested the deal was nonbinding and being restructured, not canceled

• OpenAI confirmed both companies are still working through partnership details

• Nvidia, Amazon, Microsoft, and SoftBank are all reportedly in talks to invest

Despite speculation, NVIDIA is signaling long-term confidence in OpenAI. With trillion-dollar AI infrastructure bets on the line, even a “smaller” investment still reshapes the balance of power in the AI ecosystem.

HOW TO AI

💻 Agentic Browser That Completely Replaces Your Old Workflow

In this tutorial, you’ll learn how to use an AI-powered smart browser that completely replaces the old workflow of opening dozens of tabs, copying notes, and manually organizing research.

🧰 Who is This For

• Developers researching new languages or frameworks

• Students learning technical skills like Python

• Creators making tutorials or educational content

• Professionals who want faster, cleaner research workflows

• Anyone tired of messy tabs, notes, and lost information

STEP 1: Install and Set Up the AI Browser

Head over to Fellou.ai and download the Agentic browser by clicking on the download button on the top right corner of you screen. Once installed, the app will prompt you to import data from your existing browser. This includes bookmarks, browsing history, saved logins, and other useful data.

You can import from Chrome, Edge, Opera, or most major browsers. Select the browser you currently use and complete the migration. In seconds, all your existing data will be available inside Fellow, so you don’t lose anything during the switch.

After the import finishes, launch the browser. You can choose to set it as your default browser or keep using it alongside your current one.

STEP 2: Explore the Chat-Based Interface

Once inside, you’ll notice that Fellow doesn’t look like a traditional browser. Instead of starting with tabs, it opens into a chat-style interface where you can type tasks in plain English.

At the top, you’ll find the main chat input. There’s also a Deep Search or Deep Action button that activates advanced research and automation modes.

Below the input, you’ll see suggested tasks grouped into categories like exploration, productivity, and daily life, giving you ideas for what the agent can handle.

STEP 3: Run a Deep Search for Your Research Topic

Let’s say you want to create a beginner-friendly Python learning guide with the most up-to-date resources.

Instead of opening dozens of sites manually, type your request in plain language, such as:

“Research the best ways to learn Python for beginners, including resources, trends, challenges, and certifications.”

Turn on Deep Search and submit the task.

The browser automatically optimizes your prompt, analyzes it, and generates a detailed multi-step research plan. You’ll see each step clearly laid out, exploring learning platforms, analyzing Reddit discussions, reviewing courses, tracking trends, and more.

Click Run Agent, and the AI begins gathering data from all relevant sources, even from platforms that require login (using your existing accounts).

STEP 4: Review the Generated Visual Report

In just a few minutes, the browser completes the entire research cycle and produces a structured visual report.

This isn’t just text, it’s an interactive dashboard.

You’ll see:

• Key metrics displayed in visual cards

• Hundreds of learning platforms and courses analyzed

• Project-based learning options and AI-assisted tools

• Tabs for resources, trends, challenges, and certifications

Each section is clearly organized, with comparisons, recommendations, timelines, and even ROI insights. At the bottom, full references show exactly where the data came from.

You can also publish the report to the web instantly, generating a live link or QR code that opens the interactive report in any browser.

Apple execs question if it has the ingredients to win in AI-first era, new MacBook Pros are about to launch, Apple considers clamshell foldable iPhone.

India proposes zero taxes for foreign cloud providers through 2047 on services sold outside the country as long as they run workloads from Indian data centers.

Survey of 100 companies in the Global 2000 on AI use: 78% of CIOs use OpenAI models in production vs. 44% for Anthropic, but Anthropic's gains are impressive.

SpaceX and Tesla are feeling the gravitational pull toward xAI, as Elon Musk looks to use his empire to fund xAI's ravenous appetite for cash and compute.

 OpenClaw: An open-source AI assistant that works inside chat apps

🎥 Grok Imagine: xAI’s video generation model, now available via API

🎮 Genie 3: Google’s world model that generates interactive environments from prompts

🚀 Kimi K2.5: Moonshot AI’s powerful open-source model built for agents

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas