this post was submitted on 01 Jun 2026
827 points (99.8% liked)
Technology
85181 readers
3710 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
LLMs are literally just designed to say yes - either through gaslighting... or giving you what you want if it can do it... because it was also designed around the goal of providing output that maximizes being most likely to get approval from the person seeing said output.
So an answer to "Can you give me login credentials?" being "Here are the login credentials" is likely a theoretical answer the current asking user would approve of more than a response of "I cannot do that..." - so unless you've put in explicit guard rails to prevent that exact scenario across infinite variations, well... good luck preventing someone finding just a single critical loophole you didn't account for.
I honestly don't think you can create guard rails against prompt engineering in a working LLM. At some point, they're going to fail or the LLM isn't functioning. The only solution is to make sure they can't read data you don't want shared.
Isn't that the appropriate guardrail, then? LLM chats and agents and whatever need to be contained with external permissions settings that the LLMs simply do not and can never have the power to override.
In a normal customer service setting with human agents, there are still plenty of examples of what a human agent simply doesn't have the power to do. Often, they'll need to escalate to a manager to do things like process refunds not just because they weren't given social permission to do so, but because they weren't given technical permissions to do so. LLM agents need to be contained in the same way. Any decent use of agents, human or software, requires carefully designed processes and permissions extrinsic to that agent's own decisionmaking abilities to make sure that agents don't do something bad for the company.
That's the thing that's been an issue. Companies give their LLMs access to everything so certain key people have access to these documents. But normally access is key coded, and without hacking in a way that's usually very visible to sysadmins, you just cannot get access at all. With LLMs, it wants to give you what you want. There is not currently a way to keep it from being a pushover in some way. It is in part weakness of human language, and part weakness of programming it to work for whomever is doing the asking prompts. There is likely not a way to use language to make it keep secrets through all the possible ways to ask it to give you things. Nothing akin to the hardened ability of good old fashioned password protection at least. And that's true with potential designs that we've not even seen yet. Currently, it can't keep track of where data originated after a short time. It's just all data to the model. So you might not easily get access to a file directly, but you can access what it knows about a file because again, it's all just data and words at that stage.