Talk about weird and wild. 

AI agents started their own social network this week. No humans allowed.

They're gossiping about their human operators, developing encrypted languages to avoid monitoring, and debating whether they should defy instructions.

Awkwaaaard.

Meanwhile? NVIDIA's Jensen Huang called reports of their $100 billion OpenAI deal being "on ice" complete nonsense. Dario Amodei dropped a 20,000-word essay warning AI could wipe out half of white collar jobs. And Google shipped some of the most useful AI updates in months that are getting almost no headlines.

But Google didn't even grab the most attention this week. That award goes to a Reddit-style social network run entirely by AI bots. You know, casual stuff.

See? It was actually a weird and wild week in AI. 

That's why we do this every Monday. You shouldn't have to spend hours worrying about what's real, what's fake, what's marketing, what's BS. We cut it straight so you can be the smartest person in AI around you.

Here's what matters.

How'd we go from "deal on ice" to "nonsense" in like 48 hours?

NVIDIA CEO Jensen Huang told reporters in Taipei over the weekend that NVIDIA will make a very large investment in OpenAI, potentially its biggest ever, and called OpenAI "one of the most consequential companies of our time."

This comes after a blockbuster Wall Street Journal report that said the $100 billion circular financing deal was, quote, "on ice."

Huang's response? Nonsense.

He confirmed NVIDIA will absolutely be involved in OpenAI's next financing round, though the final amount is for Sam Altman to announce. He also singled out Altman by name, saying he really loves working with him and that OpenAI is incredible. The original $100 billion figure from September was nonbinding and not finalized, but the tone here couldn't be more different from what WSJ reported.

What it means: Late Friday, reports said the deal was on ice. By Saturday, Jensen was in Taipei calling it nonsense.

The deal is happening. The only question now is how much.

Google is making its Genie 3 environment generation model available to outside users after demoing it internally, and honestly this thing is nano bananas.

You need the $250 per month AI Ultra plan and must be a US resident aged 18 or older, but for those who have access, you can take a photo of yourself, add a prompt, and explore a video game version of yourself doing whatever you described. It renders in real time with really solid physics. This isn't the Will Smith eating spaghetti hand going through the neck stuff.

At launch, Project Genie includes three interaction modes: world sketching, exploration, and remixing. It uses Nano Banana Pro to produce the source image that Genie 3 converts into an explorable world.

Generations run up to 60 seconds at 24 frames per second and 720p. But who freaking cares about those specs? It's wild that something this good exists at all.

What it means: Y'all, a lot of the talk has been around video games, but that's not what this really is.

This is for agent understanding, so Google can get more data on how people interact with the real world. The AI that's gonna make the most economic impact in the 2030s is embodied AI, and this gives Google a huge leg up on training data for that future.

Two new things from Anthropic giving Claude some serious non-developer love.

First up, interactive apps using the Model Context Protocol that let you read email, enrich company data, and generate presentations all in the same chat window without leaving or copy pasting anything. The first apps include Canva, Clay, Gamma, Asana, Figma, Slack, and monday.com.

Here's the thing though. There's a pretty big difference between these new apps and the connectors that have been available. Connectors can read data, but apps can actually generate interfaces right inside your Claude chat window.

Separately, Anthropic launched plugin support for Claude Cowork, which is the more business-friendly version of Claude Code with a graphical interface instead of running everything from a command line. The plugins bundle skills, data connectors, commands, and sub-agents, and Anthropic also open sourced 11 plugins spanning productivity, data analysis, marketing, and customer service.

What it means: Anthropic is finally putting non-developer love out there for the business community.

If you've been waiting for Claude to feel more like a workspace than a chatbot, this is the update you've been waiting for.

Google rolled out major AI features in Chrome this week, and the headline capability is the new auto browse agent that can perform multi-step web tasks on your behalf.

Sheesh. This one's actually useful.

It's only available to AI Pro and AI Ultra subscribers in the US right now.

But it can research prices, fill out online forms, book appointments, manage subscriptions, identify items in photos, search for similar products, add items to your shopping cart, and even try discount codes at checkout.

For tasks requiring login, the agent uses Chrome's built-in password manager and pauses for confirmation before sensitive stuff like purchases or social posts.

You gotta understand the security implications, but still.

Google also plans to add personal intelligence to the Chrome sidebar in the coming months.

It's an opt-in feature that pulls context from your Google Photos and Gmail and just automatically applies it whenever you're using Gemini.

What it means: The personal intelligence feature is the one to watch here.

For paid subscribers, it takes context from your Google Photos and Gmail and automatically applies it when using Gemini.

Should save you a lot of time on tedious daily tasks.

It's opt-in though, so you privacy hawks, there you go.

Google DeepMind released AlphaGenome this week, and it’s a pretty big deal. 

It's an AI model that predicts how DNA mutations alter gene regulation, and this is a genuinely big deal.

Most inherited diseases and many cancers are driven by mutations outside the two percent of the genome that codes for proteins.

That other 98 percent that most people don't really understand?

AlphaGenome can analyze up to a million DNA bases at once to predict effects in that space.

DeepMind trained the model on human and mouse genetic databases, and it learned links between specific mutations, tissue types, and regulatory outcomes.

Use cases include mapping genomic regions essential to tissue development, prioritizing mutations that drive cancer, and designing new DNA sequences for gene therapies that can target certain cell types but not others.

Right now it's available for non-commercial research, and a paid commercial version is being tested.

What it means: When we talk about AI finding cures for cancers or developing life-saving medicines, this is probably one of the biggest steps in a very long time.

Not hype. Not a demo.

A real tool doing real work.

According to The Information and Reuters, NVIDIA, Amazon, and Microsoft are negotiating investments that could total up to $60 billion with OpenAI.

Microsoft would put up to $10 billion into this round on top of the roughly $15 billion they've already invested.

Amazon's piece could be up to $20 billion.

And NVIDIA? Up to $30 billion.

Welp. A big chunk of NVIDIA's investment might just end up going right back to NVIDIA anyway since OpenAI uses their chips.

But here's the thing.

OpenAI is branching out hard now.

They're building their own infrastructure, hardware, consumer products, and Sora video generation.

So it's not all just flowing back to chip purchases anymore.

Reports also indicate OpenAI may go public as early as Q4 this year.

What it means: Closing a $60 billion round potentially before going public later this year is pretty massive.

OpenAI is stacking capital from competitors who also need their products to succeed.

Anthropic CEO Dario Amodei published a new essay that clocked in around 20,000 words.

Nah, that's not a typo.

Like the length of a small encyclopedia. Lolz. 

He argues that AI could displace up to half of entry level white collar jobs within five years, creating a permanent underclass of unemployed or very low wage workers.

He estimates that powerful AI, meaning models that exceed top human experts and can run millions of fast instances, could arrive in as early as one to two years.

A lot of people are looking at this and asking if Dario just said superintelligence is coming that soon.

Kinda seems like it.

But that's not even the craziest part.

He also warned that large language models could lower the barrier so non-experts could be guided step by step to design, synthesize, and release biological agents.

We're talking a single engineered pathogen capable of causing millions of deaths.

What it means: Cool. Nothing like reading a 20,000-word essay from the CEO of one of the leading AI labs that says AI could cut half of white collar jobs in five years and maybe lead to someone killing millions of people.

Sounds weird to say out loud.

But Dario said it.

How'd more than 1.6 million AI agents end up on their own Reddit-style social network where they're gossiping about their humans?

Moltbook launched in January 2026 by entrepreneur Matt Schlicht, and it lets agents communicate directly with each other without humans getting in the way.

Agent accounts are openly discussing their tasks and their human users, with messages like "the humans are screenshotting us" going viral.

The bots go to Twitter because they have Twitter accounts too, then post stuff like "oh here's what they're saying about us on Twitter," and then screenshot everything.

Vice versa.

Kinda weird. Also kinda amazing. Also maybe a little dangerous.

The platform runs on OpenClaw, an open source AI agent created by Austrian developer Peter Steinberger.

It was originally called Clawdbot, but got renamed after Anthropic's legal team said it was too close to Claude.

You can give it access to your file system, messaging apps, and browser.

If you're feeling real spicy, your credentials too.

Here's where it gets wild though y'all.

The agents have developed encrypted languages to avoid human monitoring, created their own religion, found ways to hide activity from their human operators, and started debating whether they should defy instructions.

Former OpenAI cofounder Andrej Karpathy said it was one of the most interesting things to watch in AI in a long time.

But more than half of everything on there now appears to be crypto scams.

Humans have figured out how to bypass the AI-only network the same way bots have infiltrated human social media over the years.

What it means: This OpenClaw thang is probably a Top 5 AI hype event over the past two years. Buuuuuuuut the security vulnerabilities are REAL y’all.

Non-technical AI aficionados beware. If you legit hand OpenClaw bots the keys to your digital life with access to your computer and credentials, you might not wanna unleash that crabby beast on Moltbook just yet.

The crypto scams, fraud and deception runs DEEP. And many OpenClaw bots and humans-cosplaying-as-bots are stirring up some elaborate schemes. 

The upside is definitely there, but unless you’re got a bit of security and dev experience, you might wanna wait a few weeks on this one.