DLP for LLM's

Mohamed Chorfa 0 Reputation points
2024-01-10T15:31:50.2033333+00:00

Hello,
Is there some sort of DLP for LLM models? For a prompt life cycle, we want to apply DLP for all prompts/completions in/out. PS: I already looked into Azure Content Safety and Azure Purview.
Thanks

Azure OpenAI Service
Azure OpenAI Service
An Azure service that provides access to OpenAI’s GPT-3 models with enterprise capabilities.
4,083 questions
Microsoft Security | Microsoft Purview
{count} votes

1 answer

Sort by: Most helpful
  1. David Broggy 6,291 Reputation points MVP Volunteer Moderator
    2024-01-10T19:12:42.17+00:00

    Hi Mohamed, I've also been watching for such a feature but I'm not satisfied with what I've seen so far.

    That said...

    Defender for Cloud Apps has a new feature for blocking unapproved llm activity: https://learn.microsoft.com/en-us/defender-cloud-apps/release-notes

    However if you're building an LLM solution I'd suggest including a logging layer so all requests are logged.

    If you can send those logs to Sentinel (eg. syslog or logic app api to your app) then you can use KQL to query for patterns of interest. Hope that helps.

    0 comments No comments

Your answer

Answers can be marked as Accepted Answers by the question author, which helps users to know the answer solved the author's problem.