Control and Manage LLM APIs with Anypoint Flex Gateway

  • Overview
  • Resources
  • Transcript
Speakers
Sue Siao
Technical Product Marketing Manager, Salesforce

Description

Technology is moving fast and the recent introduction of AI innovation is exciting, especially with the promise of increased productivity. However, the innovation comes with overall security risks of accessing data across various systems and ensuring that private data doesn't end up in a public large language model for others to potentially access.

Learn how you can manage your LLM APIs integrated into your applications using Anypoint Flex Gateway and Custom Policies with PII Detection and API Key Management.

[00:00:07.368]

Many of us are excited about A I and

[00:00:09.569]

its potential to improve our systems

[00:00:11.608]

ways of working and business processes.

[00:00:14.398]

This means that A I is about to be integrated

[00:00:16.879]

with various business critical systems

[00:00:18.920]

with sensitive information

[00:00:20.969]

and this poses several challenges for it.

[00:00:23.109]

Teams such as fragmented

[00:00:25.149]

data across growing applications which

[00:00:27.440]

leads to missing context,

[00:00:29.219]

una actionable A I insights, uncertainty

[00:00:32.189]

and what will happen to the data center LLM

[00:00:34.679]

and risk to data security.

[00:00:37.719]

In this video, we will cover how it teams

[00:00:40.029]

can ensure that these innovations with A I

[00:00:42.158]

do not end up with confidential data leaks.

[00:00:45.609]

Large language models or LL MS

 

[00:00:47.668]

are mainly accessed through API

[00:00:49.709]

S. This is what allows them to be integrated

[00:00:52.000]

with various systems and can be used by end

[00:00:54.189]

users

[00:00:55.880]

and API S are best managed by an

[00:00:57.950]

API gateway. Most commercial API

[00:01:00.240]

gateways comes with out of box policies

[00:01:02.450]

for authentication contracts and rate limits.

[00:01:05.209]

However, this does not account for

[00:01:07.430]

data present in the prompts

[00:01:10.239]

to ensure that only selective data gets sent

[00:01:12.359]

to LL MS through the prompts. We recommend

[00:01:14.500]

that first users authenticate through

[00:01:16.829]

the gateway.

[00:01:17.859]

Two limit the usage to reduce

[00:01:20.069]

unnecessary costs. Three manage

[00:01:22.418]

API keys with the LLM vendor

[00:01:24.519]

centrally

[00:01:26.028]

four check if the data sent

[00:01:28.069]

to the LLM are compliant before

[00:01:30.299]

sending the request to the large language model.

[00:01:33.370]

Any point Flex gateway from Mulesoft is

[00:01:35.409]

an ultra-fast lightweight API gateway

[00:01:37.588]

to control and secure any API deployed

[00:01:40.000]

anywhere.

[00:01:41.138]

In addition to out of the box policies, Flex

[00:01:43.489]

gateway allows your team to create custom

[00:01:45.569]

policies so that you can extend the gateways capability.

[00:01:48.888]

Let's see how any point Flex gateway can

[00:01:50.989]

control access and usage of LLM

[00:01:53.219]

API S with a combination of out of the box

[00:01:55.750]

and custom policies

[00:01:58.668]

within any point platform. We have API

[00:02:00.790]

manager where we can go ahead and control

[00:02:03.400]

and manage API S

[00:02:05.129]

here. We have open A I API

[00:02:07.659]

added to Flex gateway

[00:02:09.849]

within API manager. We can set up SL

[00:02:12.099]

A tiers contracts, alerts

[00:02:14.460]

and also set up policies

[00:02:16.500]

all through one user interface.

[00:02:19.300]

We already have authentication and rate

[00:02:21.528]

limit policies added which are

[00:02:23.618]

out of the box, but we also want

 

[00:02:25.770]

to have data protection in API key

[00:02:27.939]

management.

[00:02:30.058]

So before we start adding more

[00:02:32.419]

policies, let's make a test call

[00:02:34.469]

to the API.

[00:02:37.719]

As you could see, we weren't able to

[00:02:39.770]

make a request to open A I API and

[00:02:41.868]

it's because the user doesn't have

[00:02:44.159]

the API key information for open

[00:02:46.479]

A I. So now

[00:02:48.770]

let's go back and add the

[00:02:50.808]

API key management, custom policy.

[00:02:53.949]

So within API manager,

[00:02:55.960]

you can choose to add different policies.

[00:02:58.800]

And Flex gateway already comes with out

[00:03:00.960]

of the box policies for authentication

[00:03:03.740]

rate limit

[00:03:05.129]

Heather injection and so much more.

[00:03:07.659]

But you can also develop your own custom

[00:03:09.990]

policies.

[00:03:11.629]

The API key management custom policy

[00:03:14.058]

sends the API key to the LLM API

[00:03:16.740]

so that the end user can have access

[00:03:19.240]

without actually needing the key.

[00:03:22.110]

This means that your organization can

[00:03:24.159]

centrally manage API keys and

[00:03:26.278]

the user will still be able to interact with the

[00:03:28.319]

LLM API without any hiccup.

[00:03:31.118]

OK? Now that the policy is enabled,

[00:03:33.149]

let's make another test call.

[00:03:35.288]

So here we see the response but

[00:03:37.319]

personally identifiable information

[00:03:39.580]

or PI I was sent to the OM. And

[00:03:41.868]

we definitely don't want that. So we need a way

[00:03:44.319]

to be able to block these calls with sensitive

[00:03:46.520]

information.

[00:03:47.538]

And by using our another custom policy

[00:03:50.058]

for PI I detection,

[00:03:52.118]

this custom policy checks for pi I

[00:03:54.338]

imprompt and contact sent to the

[00:03:56.538]

LM API using an open source

[00:03:58.679]

data protection. And the ISDK

[00:04:02.330]

when private information such as credit card

[00:04:04.569]

numbers, names or driver's license

[00:04:06.649]

information is part of the prompt or the context,

[00:04:09.649]

the policy detects them and lets the user

[00:04:11.778]

know it will also block the prompt from

[00:04:13.879]

reaching the LLM API in the first

[00:04:15.889]

place.

[00:04:17.350]

If there is no sensitive data, the gateway

[00:04:19.500]

will pass the request to the LLM API.

[00:04:23.309]

So now let's see how all these policies

[00:04:25.709]

work together by making a test call.

[00:04:28.970]

The gateway responds that the request body includes

[00:04:31.290]

sensitive data and rejects the request

[00:04:34.350]

with a combination of out of the box policies

[00:04:36.730]

and custom policies. With any point Flex gateway,

[00:04:39.209]

you can proactively protect the data sent

[00:04:41.379]

to LM API S while allowing

[00:04:43.769]

your organization to innovate with A

[00:04:45.778]

I any point Flex gateway

[00:04:48.149]

is now able to control traffic to the LMM

[00:04:50.369]

API through authentication rate limit

[00:04:52.660]

API key management and PI I detection

[00:04:55.540]

with a combination of out of the box policies

[00:04:57.759]

and custom policies with any point Flex gateway.

[00:05:00.139]

Your it team can proactively

[00:05:02.149]

protect the data center to LM API S

[00:05:04.178]

while allowing your organization

[00:05:06.250]

to innovate with A I.

Start free trial