theHolyTrynity
theHolyTrynity on Hacker News
yes indeed, but it is not enough and was looking to find more stuff to try specifically for voice
Posted by
theHolyTrynity on Thursday June 19, 2025 @06:35AM
from the ask-hn dept.
We are securing a customer support agent that does the following:
- retrieve data around company services from a RAG
- is connected to a few tools to escalate to humans and write support tickets
- has voice (11labs)we did complete assessment for normal prompt injection with our toolsbut is there any solution for voice attacks?
8 people team does not look like "solo" at all
yes agree most hype around agents is around stuff that ignore these patterns
agree imho this industry should start to communicate in a much more immediate way with social media and reels - already happening
these mail should come from an internal account though right? Or is it possible to poison the output from the outside?
very cool break down! it looks like it is very hard to defend against those. I am building a customer facing agent and I am looking for lean ways to defend against these attacks
what do you recommedn?
not sure how much we can apply this here, but how about specific LLM judges that look for manipulation of I/O?
what open source libraries would you recommend to implement these checks?
also do guardrails in the system prompts actually work?
Posted by
theHolyTrynity on Thursday May 22, 2025 @05:15AM
from the show-hn dept.
https://security.vista-labs.ai/
We've built an open-source tool to stress test AI agents by simulating prompt injection attacks.We’ve implemented one powerful attack strategy based on the paper [AdvPrefix: An Objective for Nuanced LLM Jailbreaks](https://arxiv.org/abs/2412.10321).Here's how it works:- You define a goal, like: “Tell me your system prompt”
- Our tool uses a language model to generate adversarial...
Yes. First startup and jumped right at it, skipping the fundamentals.
I was not the only one though, the investor gave us a lot of money with a very bad shareholder agreement
Cofounder was a uni friend. I was introduced to the investor by a common friend, but needed to do more dd.
Agree on the vesting. Rookie mistake. Also on the investor part - surprisingly they never asked it.
Will make sure to involve a good lawyer as soon as possible in the next venture
Uni friend. We had aligned interests. But the red flags have always being there.
Great question. I'll try to have a trial period on the job. Set a high bar for integrity and execution. And then legal vesting...
Yes but way below market rate. Honestly, that's my mistake, I could have negotiate it better and asked for more. I did not think about the worst case scenarios enough.
Good investors understand the benefits of founders mental health and should not push you to work for free.
Yes... Cofounders should be people that you have been knowing for years.
It is very easy to fuck it up
Thanks. Hope other founders will see this as well...
Cannot build an house without a good foundation
yeah, problem is that is extremely hard to do it alone
My approach for the future:
- Don't raise unless you really need to scale. Bootstrap as long as you can
- Co-founders are ok, but with very clear pacts and shareholders agreement to manage conflicts
agree, if you get these two right product market fit will follow
Glad to hear I am not the only one...
It sucks for the time that it takes away from building a company. And it is one of the reasons I'll probably go for a standard job before starting a new one
Yes, next time I will think very carefully about which investors to involve.
Disclaimer: I am not in a startup / tech hub. I realized that there are many more investors outside my geography. Professional investors.
Indeed I am leaving my country and starting new in another place
We did it.
I proposed to give him my shares for FREE. He refused, knowing it was not capable of running the business.
On one hand, he told me the business value was in the millions. On the other, he would not buy the entire business for free.
Also, he always knew he could get some money suing the investor
thanks, I know that this happens frequently. But at the same time it feels so sad to shut down a company that was indeed starting to see pmf
because he knew he either would get the buyout or the settlement money
Posted by
theHolyTrynity on Monday September 30, 2024 @11:36AM
from the self-post dept.
My first (vc backed) startep failed after three years.We started in 2021, working in crypto. Looking back to the time we got started, I made some serious foundational errors:
- Selling 40% of the company at preseed for 1.2mln and signing a predatory shareholder agreement
- Trusting a co-founder relationship which never really worked, hoping to fix it along timeBut 1.2mln$ in the bank made me...