The Story I Keep Getting Asked to Tell - My new brand and AI did not play nice

MAKE SURE TO CHECK OUT THE UNBELIEVABLE THING CLAUDE SAID TO ME AS I WAS WRTING THIS ON THE PLATFORM- AT THE BOTTOM OF THIS POST. Every time I re-tell this story my body goes in to full PTSD mode, I start shaking, my whole body gets tight and clammy and my jaw clenches- so I thought I would write my story out about my negative experience with AI during launch week -

For seven months, I quietly built a brand.

Not just a logo or a website. A studio. A new identity for almost a decade of work — refined, strategic, story-driven, the kind of brand I had spent years helping other founders build, finally turned inward toward my own.

Portia & Cleo.

Every piece of it — the name, the manifesto, the visual identity, the messaging architecture, the entire systems backend — was built with the kind of care I bring to my clients. Seven months of slow, intentional work. I was finally ready to launch.

And then, in the exact same week I was supposed to be celebrating, the bottom fell out — twice.

PART ONE

The Lockout

I had been completely sold on the future of AI-powered marketing. I was running Claude — Anthropic's AI assistant in its Cowork mode — like a business partner. Building SOPs inside my clients' Google Drives. Setting up automated systems. Cleaning up my Gmail with daily briefings. Drafting responses. Catching things I would have missed. I genuinely believed I had found something better than hiring an assistant. I was thrilled.

At the same time, I was working with a backend systems strategist who was building out in my Google account to house client assets and deliverables. She was building. I was building. Both of us were running inside Google in parallel.

And then the email came.

My Google account had been flagged for suspicious automated activity. Too many bot entries. Too much movement from AI. The account was suspended.

Effective immediately.

I appealed within minutes. Forty-eight hours later, Google granted my access back. I exhaled.

I shouldn't have.

During those forty-eight hours, the failed login attempts had stacked up so high that the system would no longer send me an authentication code. I could see the account. I could not get into it. I cleared my cache. I tried different browsers, different computers, different networks, different devices.

Nothing.

What I did not know in the moment is that I was part of a much larger pattern. Throughout the spring of 2026, waves of users running AI agents through their Google accounts have been getting flagged and suspended — sometimes permanently. Security researchers have documented that Google's systems are detecting AI agent activity as botnet behavior, and that appeals rarely succeed because the platform's classifiers are reading the patterns of automation as abuse, no matter the intent.^{1, 2}

The Google account that held the entire backend system I had spent weeks building — the one designed to elevate the way my clients would experience working with me — was gone. Not because someone hacked it in the way we usually mean that word. But because the very tools I was so excited about, the agents I had set in motion on my own behalf, had set off alarms loud enough to lock me out of my own house.

In the exact same week I was launching.

PART TWO

The Hack

Once I accepted that the free account was unrecoverable, I made a decision. I would rebuild from the ground up. This time, I would pay for a Google Business account — because at the very least, a paid account meant I could talk to a human being when something went wrong.

I went through the setup. Toward the end of the flow, there was a button.

It looked harmless. It looked helpful. It offered to use Google's AI features to optimize my new Business account. I had no reason not to click it. I clicked.

What I did not know — and what only GoDaddy and Wix would explain to me later — was that the question I had just answered sat inside a known vulnerability. A vulnerability that was actively being exploited.

The moment I clicked, a hacker moved.

They swooped in and tried to take over the brand new domain I had purchased ( 2 months prior)through GoDaddy. They could not touch the website itself, because it is hosted on Wix and I had — thank God — paid for a security program that protected it. But they were able to seize five different ownership pieces of my email for my new brand name. The same brand name I had just spent seven months building.

GoDaddy had to file an appeal through Google to recover them.

Three days. Three days of phone calls, of explaining, of waiting to find out whether the brand I had built would survive the very week I was bringing it into the world.

What happened to me has a name in the security world: indirect prompt injection paired with an OAuth-level vulnerability. Security researchers have spent the last year documenting how attackers are hiding instructions inside the very AI integrations that platforms now invite users to enable — instructions the AI then carries out on the user's behalf, with the user's permissions.³

Microsoft documented a near-identical zero-click attack inside its own AI assistant, and security researchers have warned that the open web is now "filling up with traps designed for LLM-powered agents." The Google OAuth ecosystem in particular has had documented flaws that allow account takeover when domain ownership or AI-feature permissions shift hands.^{4, 5}

I am not someone who fell for a scam. I am someone who clicked a button that the platform itself offered me.

I rebuilt. I got the pieces back. I got the brand back.

But something in me did not come back the same. Loosing brand assets and cleint work was a first for me and foever changed me as and entrepreneur.

IN CLOSING

Where I Am Landing

As I sit here today, I have AI thought partners helping me write this blog. The same Claude assistant from Anthropic that was inadvertently part of what set this whole story in motion is the very one helping me find the language to share it now. I am using it to organize my stream of consciousness, to find the words for what I just lived through, to give shape to all of this on the page.

I believe in the future of AI.

I just went too hard, too fast — and I know now that there are so many vulnerabilities sitting in the gap between what AI can do and what the platforms underneath it can yet hold.

According to a 2026 QuickBooks report, 79% of women entrepreneurs expect AI to play a role in the future of their business.⁶I am one of them. And exactly because I am one of them, I owe it to the women coming behind me to be honest about what the dream is currently costing.

As a solopreneur who has built this business from the ground up, now in year five of rapid growth, I cannot afford to risk what I almost lost again. I cannot rebuild that brand a third time. I cannot ask my business to absorb another week like the one I just lived through.

So I am taking a beat.

I am going back to some old-school ways. Slower. Quieter. With more of my own hands on the work, and fewer agents in the background. I am going to write my own emails for a while. I am going to organize my own folders. I am going to let some things take a little longer than they could.

And I look forward to the future — when the security catches up, when the platforms hold, when I can fully embrace the rapid growth of AI again with the trust I once had in it.

But for now, I am going to be a little old-school.

I am still here.

Still building.

Still telling the story.

Just a little more carefully now.

AFTER NOTE-

And I had to throw this in- Because you literally cannot make this up- as I was woking on this post with Claude Co Work - I kid you not this is a response I got:

SOURCES & FURTHER READING

1.  Why Our AI Agent Got Suspended (And the Irony of What It Wrote Before That Happened)  —  Medium, April 2026

2.  What to do if Gmail bans your AI agent  —  AgentMail

3.  Indirect prompt injection is taking hold in the wild  —  Help Net Security, April 2026

4.  When prompts become shells: RCE vulnerabilities in AI agent frameworks  —  Microsoft Security Blog, May 2026

5.  AI Security in 2026: Prompt Injection, the Lethal Trifecta, and How to Defend  —  Airia

6.  Women Entrepreneurs 2026: Trends in Funding, AI, and Growth  —  QuickBooks