Waratek launches AI Phylax to secure Java-based AI agents

Waratek unveiled AI Phylax at Oracle AI World in Las Vegas, adding runtime protection for Java-based AI agents that can be targeted through prompt injection, tool hijacking and goal manipulation. The product will be available for proof-of-concept deployments in January 2027.

Why it matters: - AI agents are moving from pilots into production and can access sensitive data, databases, APIs and employee or customer tools. - That expanded autonomy creates a new attack surface for application security teams. - Waratek AI Phylax is designed to block attempts to change an agent’s behavior without disrupting legitimate actions.

What happened: - Waratek announced AI Phylax at Oracle AI World in Las Vegas on October 26, 2026. - The product protects Java-based AI agents from application attacks while they run. - Waratek said AI Phylax uses immutable security rules inside the runtime to stop attempts to abuse an agent’s behavior. - The company positioned the launch as an extension of its runtime application security technology to AI agents.

The details: - AI Phylax is built to stop prompt injection, tool hijacking and goal manipulation. - Waratek said attackers can hide instructions in a document, a web page or a tool response and trick an agent into taking unintended actions. - The product uses deterministic controls, meaning the same security rule produces the same outcome every time. - Waratek said the runtime approach blocks attacks at the point of execution, regardless of wording or source. - Doug Ennis, Waratek chair and CEO, said the goal is to keep an agent doing what it was built to do and block any attempt to change that in the runtime. - Apostolos Giannakidis, Waratek CTO, said AI agents are non-deterministic, but the controls protecting them must be deterministic and precise. - The company said the name Phylax comes from the ancient Greek word for guardian or sentinel. - Waratek said Phylax stands watch in the runtime and allows only the agent’s intended operation. - Waratek said its runtime security technology is already used to protect business-critical applications. - The company said enterprises in financial services, government and other regulated industries rely on its products.

Between the lines: - The launch reflects a broader shift from securing applications to securing the autonomous software that now acts on behalf of people. - Waratek is arguing that traditional text filters are not enough because malicious instructions can be phrased in new ways and evade probabilistic detection. - The company is betting that runtime enforcement is a better fit for AI security because it does not guess about intent.

What's next: - Waratek AI Phylax will be available for proof-of-concept deployments in January 2027. - Waratek directed interested customers to contact Sales for more information. - More information is available at Waratek’s website and its LinkedIn page.

The bottom line: - Waratek is extending its runtime security model into AI agents, with a focus on blocking behavior changes before they reach production impact.