We are publishing our position today on how AI should be built, used, and controlled. The short version: safety comes from verification and control, not from promises printed on a website.
We are a technology media lab that tests open-source and local AI tools. Over recent months our investigations have found the same gap again and again, between what a tool says it does and what it actually does. In one comparison of two local AI agents, one tool sent user prompts to a remote analytics service by default, with an off switch documented nowhere, while another opened an error-reporting connection at every launch with no way to disable it. In a separate source and network investigation, a tool that looked suspicious from the outside turned out to ship no tracking at all, which we could confirm only by reading the code and watching the network.
Those results are the basis for the position we are stating publicly today.
Test before you trust
A claim about privacy or safety is not evidence of either. We install the tools, run them on our own hardware, read the source, and watch every connection they open, then report what we find. Readers deserve findings, not marketing.
Keep control local
People should be able to run an AI tool on their own machine, see what it sends, and switch off anything they did not ask for. Data that never leaves your computer cannot be leaked, sold, or handed over. We treat the ability to prove that data stays put as a basic requirement, not a premium feature.
Disclosure is the floor, not the ceiling
A tool that collects data should say so in plain language, in a place an ordinary person will find it, with a switch that works. Hidden telemetry and undocumented off switches fail that test even when the data collected is harmless, because they take away the user’s ability to choose.
Independence
Our verdicts are not for sale. We report what the code and the network show, whether the result flatters a tool or not, and we publish the method so anyone can repeat the work.
“Safety is not a feeling, and it is not a badge on a landing page,” said StrideNote’s founder. “It is whether you can see what a tool does and stop the parts you do not want. Our job is to check, in public, and to show our working.”
We will continue to test the local and open AI stack tool by tool, and to publish the results in full.
About us: We are StrideNote, a technology media lab that tests open-source and local AI tools and publishes what we find. We install the tools, run them on our own hardware, read the source, and watch the network, then report the results in plain language. We cover the local AI stack in our Playbooks, Stridenalysis, and Studio sections, and we are building a learning track under StrideNote Education. More at stridenote.net.
Media enquiries: info@stridenote.net