apunkt/plano
1
0
Fork 0
mirror of https://github.com/katanemo/plano.git synced 2026-05-03 04:42:49 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

updating doc versions, images and cleaning up section for prompt-guard (#320)

Browse source 
* updating doc versions, images and cleaning up section for prompt-guard

* updating based on feedback

---------

Co-authored-by: Salman Paracha <salmanparacha@MacBook-Pro-261.local>
This commit is contained in:
Salman Paracha 2024-12-01 23:02:08 -08:00 • committed by GitHub
parent cadd3cdaf9
commit a0c159c9ba
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
7 changed files with 41 additions and 88 deletions
Download patch file Download diff file Expand all files Collapse all files

50
docs/source/guides/prompt_guard.rst
View file

@ -1,41 +1,26 @@
.. _prompt_guard:
Prompt Guard
================
=============
**Prompt guard** is a security and validation layer designed to protect prompt-based systems by filtering and analyzing inputs before they reach the core processing stages.
In applications where prompts generate responses or execute specific actions based on user inputs, prompt guard minimizes risks like malicious inputs, unexpected errors, or misaligned outputs.
By adding a layer of input scrutiny, prompt guard ensures safer, more reliable, and accurate interactions in prompt-driven environments.
**Prompt guard** is a security and validation feature offered in Arch to protect agents, by filtering and analyzing prompts before they reach your application logic.
In applications where prompts generate responses or execute specific actions based on user inputs, prompt guard minimizes risks like malicious inputs (or misaligned outputs).
By adding a layer of input scrutiny, prompt guards ensures safer, more reliable, and accurate interactions with agents.
Why Prompt Guard
----------------
.. vale Vale.Spelling = NO
- **Input Validation**
- **Type Enforcement**: Ensures that inputs are of the expected data types, such as integers, strings, lists, or specific formats, reducing errors from unexpected data.
- **Value Constraints**: Restricts inputs to valid ranges, lengths, or patterns to avoid unusual or incorrect responses.
- **Prompt Sanitization**
- **Prompt Sanitization via Arch-Guard**
- **Jailbreak Prevention**: Detects and filters inputs that might attempt jailbreak attacks, like alternating LLM intended behavior, exposing the system prompt, or bypassing ethnics safety.
- **Intent Detection**
- **Behavioral Analysis**: Analyzes prompt intent to detect if the input aligns with the function’s intended use. This can help prevent unwanted behavior, such as attempts to bypass limitations or misuse system functions.
- **Dynamic Error Handling**
- **Automatic Correction**: Applies error-handling techniques to suggest corrections for minor input errors, such as typos or misformatted data.
- **Feedback Mechanism**: Provides informative error messages to users, helping them understand how to correct input mistakes or adhere to guidelines.
- **Policy Enforcement**
- **Role-Based Filtering**: Customizes input validation based on user roles or permissions, allowing more flexibility or stricter enforcement depending on user access.
- **Compliance Checks**: Ensures inputs meet compliance or regulatory standards, especially in fields like finance or healthcare, where prompt outputs must align with strict guidelines.
Arch-Guard
----------
In the evolving landscape of LLM-powered applications, safeguarding against prompt attacks is crucial.
These attacks involve malicious prompts crafted to manipulate the intended behavior of the model, potentially leading to undesirable outcomes.
Arch-Guard is designed to address this challenge.
.. Note::
Today, Arch offers support for jailbreak via Arch-Guard. We will be adding support for additional guards in Q1, 2025 (including response guardrails)
What Is Arch-Guard
~~~~~~~~~~~~~~~~~~
@ -60,36 +45,19 @@ How Arch-Guard Works
#. **Pre-Processing Stage**
As a request or prompt is received, Prompt Guard first performs validation, applying any type, format, or constraint checks. If any violations are detected, the input is flagged, and a tailored error message may be returned.
#. **Sanitization Stage**
The prompt is analyzed for potentially harmful or inappropriate content, and necessary filters are applied to clean the input.
#. **Behavior Analysis**
Next, the system assesses the intent and context of the prompt, verifying that it aligns with predefined function requirements. If the prompt raises any red flags, it can be modified or flagged for review.
As a request or prompt is received, Arch Guard first performs validation. If any violations are detected, the input is flagged, and a tailored error message may be returned.
#. **Error Handling and Feedback**
If the prompt contains errors or does not meet certain criteria, the user receives immediate feedback or correction suggestions, enhancing usability and reducing the chance of repeated input mistakes.
#. **Output Control**
After input validation and filtering, the prompt is allowed to proceed to the main processing phase. The output can also undergo a final check to ensure compliance with content guidelines or role-based policies.
Benefits of Using Prompt Guard
Benefits of Using Arch Guard
------------------------------
- **Enhanced Security**: Protects against injection attacks, harmful content, and misuse, securing both system and user data.
- **Increased Accuracy**: Filters out inappropriate or misaligned inputs, leading to more accurate and intended outputs.
- **Better User Experience**: Clear feedback and error correction improve user interactions by guiding them to correct input formats and constraints.
- **Regulatory Compliance**: Ensures that prompts adhere to necessary guidelines, especially for sensitive fields, minimizing the risk of regulatory breaches.
Summary
-------