Google thought I was mining Bitcoin - mining it like it's 2018 baby. But no, wasn't doing that (again).
I was running a L1s validator test node—you know, the exact kind of blockchain infrastructure that legitimate DeFi platforms actually need. The kind that requires computational resources because that's how distributed consensus works. The kind that works really well on cloud providers...
But Google's threat detection AI apparently can't tell the difference between "crypto mining operation" and "blockchain validator infrastructure."
So they banned me.
Not just for that, mind you. The ban was a trifecta:
- Running security scans (with explicit authorisation from targets)
- Operating honeypots (literal security research)
- Running that "Bitcoin miner" (a Sui test node m8)
One day I'm building away, on GCP because that's where I like to build things, the next day? Locked out. Account suspended. No more deploying. No more scanning infrastructure. No more anything.
The Irony Wasn't Lost On Me
Here I am, building a platform to help decentralised networks secure their infrastructure, and I get flagged as a threat by another AI system that can't distinguish between malicious activity and legitimate development.
Google: confused about blockchain workloads.
Simon: trying to eliminate false positives in security scanning.
The Unbanning Process (AKA: Purgatory)
I deleted this section because it was boring. Eventually, they unbanned me. Weheey!
I got back in—but with restrictions. Some things I could build on GCP. Other things? Not so much.
Conditionally reinstated.
I was annoyed at first - just sat there looking at my computer. I realise now this was a blessing in disguise.
It looked like all those hours in the data centres in the 2000s would finally pay off. Yeah, I am that old. Am I going back to bare metal!? Probably not but still, it's an option.
The Question I Should've Asked Earlier
Sitting there, freshly unbanned and afraid to breathe wrong, I had a realisation.
Even before the ban, I wasn't being smart. I'm there burning through AI API calls like they were free. Every scan would:
- Discover ports
- Identify services
- Extract banners and metadata
- Embed everything with OpenAI
- Analyse it with a GPT
Port 22 open? Ask GPT-4 if it's SSH.
Port 443 responding? Better check with the AI if it's HTTPS.
Port 3000 with a Grafana banner? Let's spend $0.03 to confirm what we already know.
I was asking a $200 billion company's large language model to tell me things I learned in 2002.
Thousands of dollars. Thousands of API calls. Thousands of Nvidia chips spinning up to answer questions like "is port 22 usually SSH?"
Four hundred times.
Same ports. Same configurations. Same fucking validator setups across different hosts.
And I kept asking. Every. Single. Time.
The Constraint That Changes Everything
The GCP ban forced a question:
How do I operate smarter with fewer resources and less aggressive scanning?
But the real question—the one I'd been avoiding—was simpler:
Why am I re-asking questions I already know the answer to?
Your brain doesn't work like this. After twenty years of staring at security scans, you recognise patterns instantly. You see an open port configuration and you know—not because you're thinking hard, but because you've seen it before. That's not intelligence. That's memory.
What Comes Next
Getting banned taught me something valuable: constraints force innovation.
I couldn't scan aggressively anymore. I couldn't just throw compute at every problem. I had to be efficient.
So I stopped asking the AI everything.
Instead, I built a system that remembers.
In Part 2, I'll show you exactly how much money I was burning on redundant AI calls, why "asking the model" isn't the same as "being intelligent," and how a $200 Postgres instance became smarter than my entire AI pipeline.
Spoiler: Vector databases aren't about replacing AI. They're about remembering what you already figured out.
Building AI-powered security infrastructure for decentralised networks. This is Part 1 of 3 on getting banned from GCP and what it taught me about building smarter systems.
Top comments (0)