Do I need and AI Gateway

With the introduction of AI tools and frameworks, all industries and sectors are looking to exploit the powers of AI to drive innovation and create savings and efficiencies. The elements of a good AI solution are complex, based on models, data, learning and intelligence. But the easiest way to release that value is through an API. Most people are very comfortable with using and providing APIs. They will now need to re-focus on security and data control.

In this blog we will be looking at how organizations who have invested in their own AI solutions, or who have implemented 3rd party AI solutions in-house need to secure the APIs exposing their AI. In the following blog we will switch our focus to how organizations using external AI services ensure they protect themselves from misuse of AI and the dangers of sharing too much with a 3rd party AI platform - once again, via the API layer.

The exposure of the API is very important and like all the APIs businesses share it needs managing and securing. Exposing an AI API without security could give consumers the keys to the kingdom. A great AI environment is put in place to release the value of the data, quickly and more intelligently. To not secure that would be very dangerous.

An AI API has very standard security requirements, some most organizations should be using already. Authentication and authorization are building blocks for protecting all APIs, and often the culprit when API security goes wrong. AI APIs are no different, their security must be founded on sound API access control practices. There may be many ways to enforce access to the API, but standards like OAuth, OpenID are perfectly created for this use case. For more sensitive AI content you may want to add to the mix best practices described in Advanced API Security Profile from FAPI and strengthen the authentication step using standards like FIDO, MFA and one-time passwords.? As well as making sure only the right people are accessing the service we also need to ensure the AI interface is not abused by adversaries, internally and externally. Data and content filters should be enforced on the inbound requests to our AI platform. The controlling of the responses should also have rules in place. The data coming from the AI platform should be checked, validated, filtered and transformed per the requirements of each organization. Sharing too much data can be very dangerous, so limiting the data going out as a response is vital. Data could also be filtered or transformed. One API can be used for a number of client applications, for example internal and external applications. Being able to filter the responses based on the type of client, or type of user making the API request will be very powerful in ensuring service re-use, saving costs, while still enforcing access based controls on your data. Controls on the AI responses messages could also be vital in making sure that APIs are not the source of data loss from your environment. Another key protection factor for any API is a rate limit. AI APIs by definition tend to be very open by nature, you ask a question and hope for a response - you don't want too many hard rules on what can be requested. But controlling how many calls come into the system is very important. Controlling access and usage is not just a pure counting exercise, the rate-limit protection needs to be able to apply rate-limits with context, per user, per application or per partner organization. The ability to count and limit the use of an AI at its API layer can also facilitate the monetization of the AI.

There are a number of solutions for protecting the AI APIs we see exploding across our industry. The most basic would be to build this into the API itself, but as the industry has seen this is not a scalable solution and the rules around authentication, protection and filtering should be applied to all APIs in a consistent way. The API security space has been bringing this value to customers for over 20 years. This supports security, governance and visibility at scale. Using something like an enterprise API gateway enables this and more, it provides all the security capabilities you would need for protecting an API. Using an API gateway also enables organizations to centrally log and record auditing information on who is using the API and how they are using it.? This space is seeing a great deal of hype and there are a number of API proxy or API management vendors who are rushing to rebrand and convert their solutions to be able to support the APIs that AI infrastructure exposes. The term AI gateway is being used to cover the topic and describe products that can protect an AI API. The Layer7 Security Gateway is a feature rich API gateway that already provides all the functionality needed to secure an API.? Organizations who already have an API gateway should look at getting extra value from the API gateway they have invested in.?

Re-applying the skills in sharing APIs to your AI interfaces will also bring some other benefits. Caching and performance management capabilities in an API platform will allow HTTP interfaces to be shared with applications internally and externally that might otherwise need large integration costs. Messaging technologies like Kafka and AMQP can bring a whole new set of use cases to the AI world. Match that with traffic prioritization and caching and the AI world can really exploit the API platforms that are out there in new ways.? Your AI interacts with the outside world via an API. At the API layer, you control the use of your AI and solutions like Layer7 are conveniently positioned to let you apply these best practices.

As this space evolves we may see standards for AI APIs evolve, maybe based on OpenAPI specifications, but the requirement to secure APIs are here now and securing those APIs is something you can do now.