Building resilient networks for AI applications
Confidently deploy reliable and protected AI workloads
Cloud network connectivity has become more critical than ever with the rise of cloud-delivered, AI-powered applications. As more organizations start to run AI workloads on their cloud networks, many teams find they need to make changes to ensure secure and reliable services.
AI models are far-reaching—pulling data from multiple points in the network: edge, mainframe, database, and cloud. In addition, AI workloads run in numerous places, often closer to data sources or users, to reduce latency. Networks must be broad, fast, and secure to support the speed and volume of data that AI workloads need.
But how do you connect users and data reliably with AI models running in dozens or hundreds of locations? And how do you manage the complexity of networking all of those locations securely? To innovate with AI, you also need innovative security and networking.
Innovate with AI securely
Working with AI models can prove tricky for developers. Many current solutions use proprietary models without the desired control or security. Open-source models are growing in popularity but can prove overly complex for many developers.
Some solutions reduce or eliminate the need for developers to manage infrastructure and security by automating tasks, such as building in controls to meet security or performance requirements, programmatically procuring and configuring needed infrastructure at deployment, and even serving as an AI assistant to accelerate better code building or to scan for vulnerabilities in real-time. Using a secure, managed infrastructure makes AI adoption easier and simplifies the complex networking needs of AI at the edge.
Protect AI investments
Infrastructure isn’t the only security consideration. Teams need AI models that are fortified as well. It starts with ensuring models can’t divulge proprietary information. Attackers may try to manipulate AI applications to reveal data, but controlling how users interact with the model is nearly impossible. AI security must be adaptable, able to evaluate legitimate prompts submitted by users and identify any attempts to exploit the model or extract data.
Controlling access to your AI models and applications is vital to security. A Zero Trust approach provides least-privilege access, deep visibility, and persistent monitoring, all reducing exploit opportunities for attackers. Zero Trust principles are also effective against the inevitable AI-powered threats. Attackers are already using AI for password spraying and brute force attacks, as well as for more effective social engineering and phishing attacks. Context evaluation as part of your Zero Trust strategy can better detect stolen credential use.
For many, AI models are delivered as SaaS offerings, making it hard to add your own security controls to them. In addition, many AI flows involve API calls between apps, which cyber attackers increasingly target. APIs must be secured to ensure the operation of your API apps, but often, organizations have unknown shadow APIs in their environment that may not have any security. Discovering these APIs and bringing them under the protection of your security tools must be a priority.
AI solutions are likely to have numerous data sources and deployment locations, meaning effective security and networking can be tricky. Consolidation is essential to reducing complexity and increasing efficiency.
Building a secure cloud network for AI
Cloudflare and Kyndryl are converging networking and security for AI to reduce complexity while improving performance. Cloudflare Firewall for AI is an advanced web application firewall (WAF) that operates on the vast Cloudflare network to prevent data loss, exposed private data, or misuse. The solution adds a new layer of protection that will identify abuse and attacks before they reach and tamper with large language models (LLMs).
Kyndryl, a world-class global systems integrator, adds the ability to implement AI security solutions and help customers understand natural language prompts, AI-assisted attackers, and the complexity of securing AI. Together, Cloudflare and Kyndryl can help you develop and deploy innovative AI and security on a trusted global network through a combination of technology and advisory services.
This article is part of a series on the latest trends and topics impacting today’s technology decision-makers.
Dive deeper into this topic.
Learn more about the future of secure cloud networking in the The Need for Network and Security Convergence whitepaper!
Authors
Ben Brillat — @brillat
VP, Kyndryl Consult Network and Edge Center of Excellence
Trey Guinn — @treyguinn
Field CTO, Cloudflare
Key takeaways
After reading this article you will be able to understand:
How AI has changed the cloud landscape
The ways in which security must adapt to better protect AI investments
Strategies to reduce complexity and improve performance