Hi Ben. Lawyer, geek and law librarian here. Also beta tester of AIs since 2017.
The irony in LLMs respecting and obeying legal fine print, is we all know their bots scrape anything irrelevant of the web sites' TOS. LegalPWN proves at least those guys are perfectly able to program their bots to understand and obey TOS ...
Hi Ben. Lawyer, geek and law librarian here. Also beta tester of AIs since 2017.
The irony in LLMs respecting and obeying legal fine print, is we all know their bots scrape anything irrelevant of the web sites' TOS. LegalPWN proves at least those guys are perfectly able to program their bots to understand and obey TOS ...
Also, Ben, aside Pangea, are there other sources confirming this ?
Not that I know of
I haven't seen additional research confirmation of LegalPwn specifically, but prompt injection in general is the OWASP #1 threat to LLMs in 2025. https://genai.owasp.org/llmrisk/llm01-prompt-injection/
Wow…language as a weapon for LLMs; they process instructions, there is no semantic meaning…