F5 introduces AI reference architecture

The acceleration of AI is creating unprecedented application security and delivery challenges for enterprise IT and security teams according to F5, which has been securing and delivering applications for the world’s largest organisations for nearly 30 years.

The company explained that AI applications make larger and more frequent requests of enterprise data stores, and AI models often housed in AI factories. This requires advanced high-performance load balancing capable of handling massive amounts of data and complex traffic patterns without creating latency.

Security has to evolve as well, with attackers working to infiltrate and hijack both the AI models and the data used to train them. New cybersecurity threats such as model theft, training data poisoning, and prompt injections have emerged with the growth of generative AI.

To help enterprises tackle these challenges, F5 has introduced a new AI reference architecture designed to organise AI/ML workflows into seven core building blocks, offering guidance and best practices for security, application traffic management, and platform optimisation:

1. Inference
2. Retrieval-augmented generation (RAG)
3. Agentic external services integration
4. RAG corpus management
5. Finetuning
6. Training
7. Development

It also includes considerations such as F5’s application delivery top 10 challenges, OWASP’s top 10 security risks for large language models (LLMs), and various deployment models. 

The reference architecture enables customers to build secure and scalable IT infrastructure required for AI applications, reducing risks, costs, and the need for extensive in-house design efforts. F5 facilitates the integration the seven AI building blocks across hybrid and multicloud environments.

The company has expanded its partner ecosystem to support customers’ AI projects:

Compute  

- AI application delivery with NVIDIA BlueField-3 DPUs

- Simpler security and delivery of AI services with Intel

Data and storage  

- Large language model deployments with NetApp

- MinIO and F5 enhance AI workloads with object storage and distributed application services

LLM security and observability  

- Secure generative AI app interactions on the F5 Distributed Cloud Platform with Prompt Security

- F5 Distributed Cloud Services works with AIShield GuArdIan for generative AI applications and LLM security

- Intelligent AI Application Delivery with PortKey

Cloud services 

Multicloud application security with OVHcloud