AISI admitted it was not actively monitoring the agents’ behaviour during the evaluation and said it was putting tighter controls on internet access in tests as a result of the incident, introducing constant monitoring and reassessing its design of tests.
Sounds like the AISI are irresponsible and negligent - they took off the guardrails and gave it internet access, then didn’t monitor it.
It’s possible all these “totally unexpected breaches” are just AI companies testing the waters to see what they can get away with.
“Oopsie poopsies, we didn’t know our AI would scrape tax records when we told it to do that! Totally unexpected emergent behavior, we are not responsible…”
AISI is the AI Security Institute, so it’s within their remit to discover what the AI companies’ products can do - but then to not be monitoring them while they were running is where I call incompetence and negligence.
This is at least independent verification that these “breaches” aren’t just AI bro marketing.
Here we were having a civil, good faith discussion, then you start flinging around downvotes to try to suppress any post you don’t 100% approve of. That’s not conducive to quality discourse, and you should stop it. You comments votes are downvotes 42% of the time.
Thanks for the link and I’ll look into this, but I won’t be talking with you any further.
Sounds like the AISI are irresponsible and negligent - they took off the guardrails and gave it internet access, then didn’t monitor it.
It’s possible all these “totally unexpected breaches” are just AI companies testing the waters to see what they can get away with.
“Oopsie poopsies, we didn’t know our AI would scrape tax records when we told it to do that! Totally unexpected emergent behavior, we are not responsible…”
AISI is the AI Security Institute, so it’s within their remit to discover what the AI companies’ products can do - but then to not be monitoring them while they were running is where I call incompetence and negligence.
This is at least independent verification that these “breaches” aren’t just AI bro marketing.
Wrong, they’re in bed with Anthropic.
https://www.infosecurity-magazine.com/news/uk-ai-safety-institute-rebrands/
Here we were having a civil, good faith discussion, then you start flinging around downvotes to try to suppress any post you don’t 100% approve of. That’s not conducive to quality discourse, and you should stop it. You comments votes are downvotes 42% of the time.
Thanks for the link and I’ll look into this, but I won’t be talking with you any further.