Thwart your robot overlords?
When man first etched a message into stone, I’m quite certain he never dreamt that one day the stone may try to read it. The surveillance and data collection within today’s information platforms would make Orwell shudder. Agentic AI is rapidly gaining traction, and it will need to “look over your shoulder” to orient itself with the task at hand.
Just earlier this year, Microsoft rolled out its Recall feature for its Copilot+ PCs. This feature would screenshot your display periodically without requiring an explicit opt-in. This presented a security nightmare. Recent testing shows that, despite Microsoft’s efforts to filter out sensitive information like passwords or credit card information, they have yet been unable to do so. Source: https://www.pcgamer.com/gaming-industry/new-report-alleges-microsoft-recall-is-still-screenshotting-credit-card-numbers-and-passwords/?utm_source=chatgpt.com
I expect to see data-mining features like this peppering our digital landscape very soon. After all, this data is useful to someone trying to understand how their systems are used! That’s going to be step one for agentic AI development.
Displays have generally been given a lot of trust in computing. At times, they have been allowed to display encryption keys and passwords in plaintext with the understanding that its content is For Your Eyes Only. This was never really a good assumption to make, but it’s brought us this far without too much trouble. Most of us were not concerned about TEMPEST vulnerabilities and the like.
The days of being able to assume that information displayed to us in plaintext has any modicum of privacy are foregone. It seems like the big players in tech are aware of this. But, I have yet to see anyone try to implement a solution that isn’t just leveraging clever RegEx somewhere behind-the-scenes.
I believe I have a completely novel idea for displaying text to a human reader in a way that is not vulnerable to single-frame capture. Given that this entire concept of displaying information in a human-readable format while obscuring it from machines is completely emergent and novel, I’ve taken the liberty of dubbing it “noography,” as it is to be inherently interpretable only by human cognition. This idea is really a sibling to steganography and cryptography, but differs in its goals. I think it is likely that there are many brilliant minds that can build on this concept to be used for similar purposes.
Here is the concept in a nutshell: generate pseudo-randomly-generated noise textures. Then, apply a single, stationary texture generated by your algorithm to the background. Apply more textures generated by the same algorithm to the foreground of the image as an animation. One may find the animated foreground reminiscent of TV “static.” Color is optional, and the background and foreground can be inverted as far as which is animated; but, I found both adding color and animating the foreground to be the most legible combination. In doing so, a single frame’s text is indistinguishable from the background as it has essentially been visually encrypted using the same key. However, the text becomes apparent to a human viewer once the image is animated. The human mind seems to be able to differentiate between the movement and lack thereof well enough to read the message. However, the message can not be deciphered from any single frame.
As they say, a picture is worth a thousand words – so here is an example. Go ahead, screenshot it! I dare you!
