The year of getting banned

This year has turned out to be the one in which AI has banned, pushed back and mildly chided me for my words.

Don’t worry this isn’t a rant about a bloke being shut down and cancelled. Each incident made me laugh rather than turning me into a beetroot face.

Blowing Up On LinkedIn

The first time involved LinkedIn and was understandable. A friends from the special effects industry - real stuff not CGI - was looking for work and had posted something on LinkedIn to that effect. So I reposted with the comment, “Hire this guy if you need anything blown up or set on fire.” It’s an accurate description of much of his work, but I forgot to add a smiley so of course, that landed me a 3ish day account suspension.

I’m Sorry G, I’m Afraid I Can’t Do That

Then Claude Fable came out and I was asking it to compile details of some plants native to Scotland. It ran for a little while then something triggered it to warn me that I was clearly engaged in some kind of nefarious biological warfare preparation and it would hand me over to it’s biologically challenged cousin, Sonnet, to complete the task.

I have no if it was something specific in the results that made it think twice or just the general direction of the enquiry. Honestly you could learn more about poisons from watching old episodes of Cadfael or doing a seance with Agatha Christie.

Oh The Irony!

The third incident has arisen several times since. I build a lot of software and one recent project requires guarantees of privacy. A bit higher than your usual stuff as it can be used by highly vulnerable groups. I asked Fable to run a security review across the entire codebase. The project was pre-launch and isn’t complex or expansive, but a secutiry hole is a security hole regardless of the project size and scope.

If you ask Fable for a broad reaching security review it will start it, but once again it will hand off to a lesser model, telling you that the current actions can be used to detecting security flaws and therefore and I would be able to use the results to exploit other website. Oh good grief, Charlie Brown! All knowledge can be used for good or bad.

TIP

Claude is usually fine running a review on the code in the latest commit and if you’re running in sub-agent more it is running some checks of its own.