CIRCLE WITH A DOT

@kbob I used to! It was a lot of fun but a lot of work for.. fun!

The fear of not using these tools is so high it’s unbelievable. I’m eventually going to have to come up with guidelines for safe use. I have legitimately more important things to do so hopefully the fervor will die down a bit by the time we re-evaluate but it’s not likely to any time soon.

If it was ours and trustworthy and reliable it would be absolutely amazing to have tools like cowork. It’s Star Trek.

I had to put on my #cto hat yesterday and reject the use of Claude Cowork and Claude Chrome for its massive security issues. Prompt injection, vm escapes (on anthropic’s side) and improper file access in the desktop app were only parts of it. The lack of auditing, transparency of every action and the fact you have to monitor its every move because LLMs are … what they are, is the rest of it.

I’m preparing an #AI Risk assessment for work. What #infosec publications or academic papers should I be looking at? Claude Cowork is top of mind.