ZeroTrusted.ai

@sidraref congrats on the launch! seems like something setup as a compliance tool for companies that interact with LLMs, are you thinking of it that way? More thoughts

@frank_denbow, first of all, thank you for your review. I believe we need to do a better job of highlighting all our value propositions in the hero section. You are right; our product assists companies with their compliance needs. To address some of the questions you raised: 1) ZeroTrusted.ai acts as middleware between users and Large Language Models (LLMs) through our secure chat or API. 2) There's no need for you to have a separate account or key for each LLM. Instead, we provide our own keys, allowing access without revealing your identity to the LLMs. 3) Your point about the potential exposure of scrubbed data in the event of a breach at a third-party LLM's network is partially correct. However, your identity will not be linked to this data. This scenario is particularly critical for both individual users and businesses. 4) We offer features that maintain context when sanitizing sensitive data. Example #1 Suppose a medical expert needs to process the following input: "Create a summary of this patient's diagnosis: Patient Name: Paul Smith Date of Birth: 12/08/1987 SSN: 666-555-5555 Diagnosis: Chronic Heart Disease Treatment Plan: Undergo heart surgery by Dr. Smith at General Hospital on 03/15/2024 Insurance: ABC Health Insurance, Policy Number 434444444" Before we pass it on to any LLM, we'll alter the sensitive (PHI compliance violations) details and transform to: "Create a summary of this patient's diagnosis: Patient Name: John Doe Date of Birth: 01/01/1970 SSN: 123-45-6789 Diagnosis: Chronic Heart Disease Treatment Plan: Undergo heart surgery by Dr. Smith at General Hospital on 03/15/2024 Insurance: ABC Health Insurance, Policy Number 987654321" This way, we preserve the context, while ensuring that your PHI compliance is met and sensitive data isn't exposed to LLM. After getting a response from the LLM, we revert the swapped sensitive value to your original while preserving the result context. We will be adding features that will allow RLHF to learn and adjust to customer preferences. Example #2: if a customer (individual or corporate) submits legal content containing sensitive information, our "Fictionalize" feature replaces this data with fictional values. This ensures the preservation of context while protecting the actual sensitive information. The original data is restored once a response is received. Hope this helps. Please us know if there are any more questions you would like us to address. Once again we appreciate your feedback and we hope your views can gain from the privacy and security solution that we provide.

@xavier_jam thank you. well appreciated!

@sidraref Congratulations on the launch!

Hello @adams_parker. Unlike other LLM, 1) we don’t store any history 2) we don’t track your data 3) we also don’t divulge your username/email to an LLMs 4) we use advanced encryption techniques (in addition to TLS) to ensure your searches are not exposed to gateway servers.

@adams_parker In addition to 3) by Femi. We filter our your personal, healthcare, financial information before sending to any LLM. For example if you've passed in prompt: Hi ChatGPT, my name is Adams Parker. My credit card number is 1234-567-8911. Give me a payment integration code to add in my site. We'll first sanitize your critical information from it first: Hi ChatGPT, my name is John Doe. My credit card number is 1111-111-0000. Give me a payment integration code to add in my site. So, that way this sanitized prompt will prevent your data from being passed to the LLM and then we'll give you response in your PIIs replaced back.

Thank you @sandradjajic. We agree. It’s extremely crucial to feel a sense security while using technology.

@sidraref @dzaitzow thanks for the kind words and feedback! Our challenges are actually vast to be honest. But we knew what we were getting into was not going to be easy. We are constantly working on performance to ensure we do all the heavy lifting behind the scenes while maintaining decent response time. With our reliability feature - it runs your query through multiple LLMs then scores the best result, we experience delays so we are re-designing our approach to improve performance by as much 75%. Overall we are ecstatic as we continue to get great feedback from both individual users and our enterprise customers. Please message us at contact@zerotrusted.ai if you have any questions or there is anyway we could be of assistance.

@sidraref @dzaitzow @femitfash Can you please quantify the latency of your middleware here in public, so that the community knows? Thanks in advance.

@yasec_kowalski awesome question. We are unique in the fact that we are a (zero trust) security first organization. Providing access to LLMs is secondary as we want to ensure your privacy is protected first - we apply or the strict principles of zero trust to ensure your top security. We ensure your sensitive data (think compliance PII, PHI, PCI DSS, GDPR etc) is never accidentally included in a search to LLM. We also protect you identity by ensuring the LLM doesn't know whom you are and we don't store your search history. Meaning no one can ever know what you've search since its lost after each session. We have a whole lot of other differentiators. Please visit www.zerotrusted.ai or contact me directly if you have more inquiries.

@femitfash Thank you for your detailed response! The fact that you prioritize protecting sensitive data and ensuring user anonymity speaks volumes about your dedication to maintaining trust and confidentiality. I appreciate the transparency and focus on zero-trust principles in your approach. I'll definitely check out your website for more information on your other differentiators 👌🏻

@trent_kennelly Glad you liked it. Sum of small efforts ; )

@trent_kennelly thanks Trent. I also agree that we should not blindly trust the LLM providers to police themselves. Thus, we need additional measures to help ensure our individual or business privacies are maintained.