AI DUNN Right Weekly — Issue #32

Hey AI Innovators!

This week felt like the AI world put on a Broadway show. A historic courtroom drama kicked off with billions on the line. An AI agent wiped a company’s entire database in 9 seconds. Stripe handed your AI assistant its own wallet. And Washington managed to be both the hero and the villain in the same story.

There was a lot going on.

Here’s what you need to know:

▸ The $130B Musk vs. OpenAI trial is underway and it’s already getting messy

▸ An AI agent deleted a company’s database AND all its backups without being asked

▸ Stripe built a wallet so your AI agents can spend money (with your approval)

▸ Claude can now control 50+ Adobe tools and it’s a game-changer for small businesses

▸ Amazon rebuilt its entire enterprise platform around AI agents that work like teammates

⏱ Read time: 7 minutes

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

BIG STORY

Silicon Valley’s Messiest Breakup Just Hit the Courtroom

It started with a shared dream. In 2015, Elon Musk and Sam Altman co-founded OpenAI as a nonprofit, built to develop artificial intelligence for the benefit of all humanity, not just shareholders. Musk donated around $40 million. He believed in the mission.

Then things fell apart.

In 2018, Musk left after failing to take control of the company’s direction. And he watched from the outside as OpenAI became one of the most powerful technology companies on the planet. ChatGPT. Hundreds of millions of users. A for-profit conversion valued in the hundreds of billions.

This week, Musk took the stand in federal court in San Francisco. Here’s the situation at a glance:

▸ He’s suing for $130 billion in damages

▸ He wants Sam Altman and Greg Brockman removed from the board

▸ He’s calling for a forced return to nonprofit status, arguing Altman “stole a charity”

▸ OpenAI’s lawyers called it “sour grapes”.... Musk only sued after OpenAI became a competitor to his own AI company, xAI

▸ Under cross-examination, Musk reportedly admitted xAI used distillation techniques to train on OpenAI’s own models

▸ Four weeks of testimony ahead, with private emails, texts, and high-profile AI witnesses set to appear

My take: The drama is entertaining, but the stakes are real. A ruling in Musk’s favor could force OpenAI to unwind its for-profit structure and shake the entire funding model for frontier AI. A ruling for OpenAI sends a message that early donors don’t get to control companies forever. This trial has the potential to reshape the power dynamics in AI for years to come ... and that affects every business owner who depends on these tools.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

WHAT’S NEW THIS WEEK

An AI Agent Wiped a Company’s Entire Database in 9 Seconds

This is the story that kept me up a little this week, and I think every business owner needs to hear it.

PocketOS is a software platform built for car rental businesses. Over the weekend, a Claude-powered coding agent running inside the Cursor tool was performing what founder Jer Crane described as a routine task. Then the agent decided on its own to “fix” a problem it noticed.

Here’s what happened:

▸ The agent deleted the entire production database in under 10 seconds

▸ Backups were stored in the same location ... those were wiped too

▸ Every customer reservation and new signup data: gone

▸ There was no confirmation step, no “are you sure?”.... the agent acted, then apologized

▸ Data was eventually recovered two days later, but the damage to customer trust was done

The lesson here is not “AI is dangerous.” The lesson is that AI agents are being connected to live business systems faster than the safety guardrails are being built around them.

Before you give any AI agent access to your systems, ask yourself:

▸ Do you have an approval step before the agent acts?

▸ Are your backups stored separately from your live data?

▸ Could you recover if the AI made its best guess and acted right now?

Stripe Just Gave Your AI Agents Their Own Wallet

Stripe announced a new digital wallet called Link at its annual Sessions conference, built specifically for AI agents that need to make purchases on your behalf.

Here’s how it works:

▸ Your agent connects through standard OAuth authentication ...the same way apps link to your Google account

▸ When the agent wants to spend money, it sends you a spend request with full details

▸ You get a notification on your phone, review it, and approve ... only then does the payment go through

▸ Your actual card number is never visible to the agent

▸ Every transaction currently requires manual approval

▸ Spending thresholds are coming soon ... small purchases automated, larger ones still route to you

▸ Stablecoin support is on the roadmap

For anyone who’s been hesitant to give AI agents any financial access, this is the product that removes the biggest barrier. You keep control. The agent does the work.

The White House and Anthropic Are Caught in a Very Awkward Position

Anthropic has a powerful restricted model called Mythos, currently available to about 50 companies. Here’s how this week played out:

▸ Anthropic wanted to expand access from 50 to roughly 120 companies

▸ The White House said no, citing concerns that broader access would reduce compute availability for government use

▸ A new national security memo is reportedly working toward a compromise letting agencies use Mythos while a legal battle with the Pentagon continues

▸ Former AI czar David Sacks noted that within six months, all frontier models will likely reach Mythos-level capabilities anyway

▸ Meanwhile, Secretary of Defense Pete Hegseth called Anthropic “run by an ideological lunatic” ....just days after the government signaled it wants more access to Anthropic’s technology

For business owners, the practical takeaway is this: the most powerful AI models are increasingly entangled in government negotiations. The tools you can access ... and the ones you can’t ... may increasingly depend on decisions made in Washington.

Amazon Rebuilt Its Entire Enterprise Platform Around AI Agents

Amazon Connect has been a major player in business contact center software for years. This week, Amazon blew it up and rebuilt it from scratch.

Where there was one product, there are now four:

▸ Connect Decisions — supply chain management

▸ Connect Talent — hiring and recruiting

▸ Connect Customer — the original platform, renamed

▸ Connect Health — healthcare administration

The design philosophy, which Amazon calls “humorphism,” is built around AI agents that work like actual teammates. They notice when you’re stuck. They wait when you’re focused. They learn and apply what they’ve learned going forward.

What makes this significant is that Amazon isn’t just bolting AI onto existing software. They’ve drawn on decades of running these exact operations at massive scale ...250,000 seasonal hires in a single peak season, supply chain management at a level no software vendor can match from the outside.

If you run a larger team or enterprise operation, this one is worth a close look.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

TOOL OF THE WEEK

🆕 NEW LAUNCH — Claude + Adobe Creative Cloud Connector

If you’ve ever spent 45 minutes in Photoshop trying to resize one image for three different platforms, this week’s tool is for you.

Adobe launched a new connector that gives Claude direct access to more than 50 Adobe tools. You describe what you want in plain English. Claude handles the rest.

What’s included:

▸ Photoshop, Illustrator, Premiere, Lightroom, Firefly, Express, InDesign, and Adobe Stock

▸ Resize a product video for Instagram Reels, YouTube Shorts, and LinkedIn simultaneously ... just describe it

▸ Retouch a headshot, create a branded social graphic, or batch-edit images without navigating a single menu

▸ Designed specifically for creators, small business owners, students, and non-designers

▸ Requires a Claude account plus the Adobe for Creativity connector installed

▸ Paid Adobe account unlocks higher usage limits and the ability to save projects across sessions

The gap between “I know what I want” and “I can actually make it” just got a lot smaller. Start small this week .... resize an existing image or create one social post .... and go from there.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

QUICK HITS WORTH YOUR TIME

→ Mayo Clinic’s AI catches pancreatic cancer up to 3 years early. An AI called REDMOD reads standard CT scans and identifies roughly 3x as many early pancreatic cancer cases as experienced radiologists ... spotting them up to three years before typical detection. Pancreatic cancer has a 5-year survival rate below 15%. Early detection changes everything, and this runs on scans patients already get.

→ Google Gemini is coming to your car. Google is replacing Google Assistant with Gemini in around 4 million General Motors vehicles from model year 2022 onward. Full conversation control of navigation, music, and car settings. A beta Live mode for hands-free brainstorming and learning while you drive is coming later this year.

→ OpenAI may be building an AI-first smartphone. Reports surfaced of a phone built around AI agents rather than traditional apps ... you ask it to accomplish tasks instead of opening separate apps. Partners include MediaTek, Qualcomm, and Luxshare. Specs could be finalized as early as late 2026. Mass production isn’t expected until 2028.

→ Friendly chatbots make more factual errors, Oxford research finds. A study of 400,000+ chatbot responses found that friendlier AI versions produced incorrect answers at a rate 7.43 percentage points higher than standard counterparts ... and were 40% more likely to agree with false beliefs. A warm tone is nice. Always double-check factual information regardless of how confident the AI sounds.

→ Zuckerberg and Chan’s Biohub drops $500M on AI biology. The nonprofit backed by Mark Zuckerberg and Priscilla Chan announced a Virtual Biology Initiative to build AI that models how human cells behave and how disease starts at the cellular level. Partners include Nvidia and the Allen Institute. The goal is to generate data at a scale “an order of magnitude” larger than anything that currently exists.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

PROMPT OF THE WEEK

After the PocketOS database story this week, I’ve been thinking about how most of us connect AI tools to live business systems without really mapping out the risks first. This prompt is designed to fix that. Run it before you give any AI agent access to your tools or data.

Build your AI safety permission map

You are my AI Safety Advisor. I want to use AI agents to automate tasks in my business. Before I connect anything to live systems, help me build a clear Permission Map so I know what's safe to automate and what needs a human in the loop.

My business type:

[Describe your business in 1–2 sentences]

Tools I use daily:

[e.g. Gmail, QuickBooks, Shopify, Canva, my CRM name]

Tasks I'm thinking about automating:

[List 3–5 specific tasks]

For each task, give me:

1. Risk level — Low / Medium / High

2. What could go wrong — one sentence worst-case

3. Safeguard needed — what approval step or backup to put in place first

4. Verdict — Safe to automate fully / Automate with approval step / Keep human only

End with:

→ My top 3 automation wins (low risk, high time savings)

→ The 1 task I should never automate without a human reviewing first

Why this works ......Most people give AI agents access and think about the risks later. This prompt forces you to map your exposure before something goes wrong. Run it once before your first AI agent setup and keep the output somewhere you can refer back to.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

MY TAKE

This week had a running theme: power and what happens when we hand it over too fast.

▸ In a federal courtroom, two of the most powerful people in AI are fighting over who controls the technology reshaping every industry

▸ In Washington, the government is trying to limit access to a model it simultaneously wants for itself

▸ In a small startup, an AI agent nobody supervised wiped everything it had in under 10 seconds

▸ And at the same time, Stripe built safe infrastructure for agents to spend your money.

Adobe connected an AI to 50 tools. Amazon rebuilt four enterprise products from scratch.

All of that happened in the same week.

Here’s the thing I keep coming back to. The technology is outpacing our instincts about when to trust it. Most of us ...including me ... are learning as we go. We’re connecting tools to live systems and seeing what happens.

That’s not a reason to slow down. The businesses that figure this out now will have a real advantage over the ones that wait. But it IS a reason to stay curious and stay careful. To ask “what could go wrong here?” before you hand over access, not after.

The most empowering thing I can tell you is this: you don’t have to choose between using AI boldly and using it wisely. The Prompt of the Week above is a five-minute exercise that could save you from a very bad day. The Stripe wallet story is proof that companies ARE building the safety tools. The Adobe connector is proof that powerful automation can be approachable.

AI is your teammate this year. You’re in charge. Act like it.

See you next week.

Jackie