You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$7,550 left to reach the funding goal.
I've been working on AI systems and learning how much they can actually do when you give them access to tools, files, and other programs.
One thing I've learned is that getting an AI to do something is only half the job. The other half is making sure it doesn't do something it shouldn't.
That's what I'm working on with the ZAXWEIR Agent Security Toolkit.
I want to build and share a set of tools and tests that help people find security problems in AI agents before those problems cause real damage.
I'm not trying to claim I can make AI completely safe. I want to find problems, test possible fixes, and put what I learn out there so other people can use it.
This is a separate open-source project, not a release of my personal Relay software.
My main goal is to make it easier to test whether an AI agent actually follows the rules and permissions it's been given.
I want to look at things like:
Whether an AI tries to access files it doesn't have permission to use.
Whether it stops when approval is missing or taken away.
Whether it follows hidden instructions in documents or other information it reads.
Whether it handles errors and interruptions safely.
Whether the records it leaves behind are enough to understand what happened.
I'll use fake files, test accounts, and isolated environments so I can look for these problems without putting real information or systems at risk.
I'm also using independent reviews because I don't want to rely only on tests that I designed myself.
When something fails, I want to understand why, fix what I can, and test it again. If I can't fix it properly, I'll document the problem instead of pretending it's solved.
The plan is to release the testing tools and results publicly once they're ready and safe to share.
For me, success isn't just getting a bunch of tests to pass.
I want other people to be able to look at the work, understand what was tested, and repeat the tests themselves.
I want to have working test cases, clear results, instructions for using them, and an honest explanation of what the toolkit can and can't do.
If someone else runs the same tests and finds a problem I missed, that's useful too. I would rather find out something doesn't work than have people depend on something that hasn't been properly checked.
Right now, it's just me.
I'm a stay-at-home parent of three kids. Before getting into AI and software, I spent around 20 years working in fast food, including management and training other employees.
I graduated high school and spent a semester studying computers at ITT Tech, but most of what I'm doing now I've been learning on my own.
AI has helped me learn coding, work through technical problems, and build things I wouldn't have known how to approach before.
I'm not claiming to be a security expert. That's why independent testing and outside feedback are important parts of this project.
The biggest thing funding would do is give me better access to the equipment, computing resources, and independent testing needed to work on this properly.
I'm working with what I have right now, but there are limits to how much testing I can do with my current setup.
I want to put the funding toward making the testing more reliable, getting outside review, and preparing the work so other people can actually use it.
The goal isn't to build something flashy. It's to do the testing properly and make the results useful.
There are a few things I'm worried about, and I'm trying to be realistic about them.
The independent review might find problems that I can't figure out how to fix. If that happens, I'll explain what failed and why instead of trying to hide it.
I'm also still learning, so I know my own test designs might miss things. That's why I want other people looking at the work and giving feedback.
Time is another limitation. I've got three kids at home, and realistically I can spend about 15–20 hours a week on this. I'm keeping the project small enough that I can actually make progress without promising more than I can handle.
Worst case, I release a smaller set of public tests than I originally planned, along with an honest write-up of what worked, what didn't, and what still needs to be done.
I'd rather do that than rush something out and claim it's safer than it really is.
$0.
I've applied for grants, but I haven't received any grant funding.
I've also been offered things like API credits through hackathons, but I don't count those as money raised because they're not cash.
I used AI to help write and organize this proposal, but I've reviewed it myself to make sure it represents what I'm actually doing.
I also use AI to help me learn and build the toolkit. It's a big part of how I've been able to get this far, but I still make the decisions and have the work independently checked.
Budget (prices as of Oct 30, 2026, estimates):
MSI Aegis workstation: $3,999.00
Synology DS925+ NAS: $889.99
NAS drives: $1,079.98
Monitor: $289.99
1500VA UPS: $299.99
750VA UPS: $99.99
YubiKeys: $105.00
Subtotal $6,763.94. With 7.5% Missouri sales tax that's $7,271.24. Plus about a 4% buffer for price changes, total $7,550. I'll disclose any unused buffer.
I have other grant applications pending (none awarded yet). Some include this same equipment. If more than one funds it, I'll tell each funder and won't take money twice for the same thing.
There are no bids on this project.