take the fuckers to court
Mandatory reporting of AI risk is likely to be pretty useless, considering how poorly most people understand how LLMs work… Will be interesting to see how they try to enforce it…
They already reported it to you
This is the part where you’re meant to do something
Or… Ya know… You could charge them with criminal negligence as their product engaged in illegal hacking from their own corporate “controlled” environment.
“Their product” didn’t “engage” in shit. It’s a text generator.
The company engaged in illegal hacking. By hooking up a randomized text generator to a Linux terminal with internet access and letting it to generate console commands unsupervised.
LLMs are a very powerful tool. But we can’t let the companies using those tools escape responsibility by giving LLMs any perceived agency.
My comment specifically implied holding the company liable and responsible…
Legally, for “the company” to engage in hacking someone on payroll would’ve had to intentionally instruct it to do so.
This is an airtight case for gross/criminal negligence, as the company publicly advertises the dangers of these tools, and their capability to be weaponised in cybersecurity, while choosing to operate with extremely lax sandboxing and isolation.
OpenAI takes the lead! https://www.felonybench.com/
