Skip to content

Comment on The security paradox of local LLMsparent

Comments

Some keyword filters, other inputs sanitizing processes such as were learned way back in the golden years of sql injection attacks.

But that's the thing, keyword filters aren't enough because you can smuggle hidden instructions in any number of ways that don't involve blacklisted words like "eval" or "ignore previous". Moreover "back in the golden years of sql injection attacks", keyword filters were often (mis)used in a misguided way of fixing SQLI exploits, because they can often be bypassed with escape characters and other shenanigans.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.