TrojAI Expands AI Security With Agent Led Protection

As enterprises accelerate the deployment of autonomous systems across development and operations, securing these environments has become increasingly complex. TrojAI’s latest update introduces new capabilities aimed at strengthening enterprise AI security by extending protection beyond prompt level controls into full agent lifecycle visibility and enforcement.

TrojAI, a platform focused on securing artificial intelligence systems, has announced a set of enhancements designed to address emerging risks associated with agentic AI. The new capabilities include agent led red teaming, runtime intelligence, and real time protection for AI coding agents, reflecting a broader shift toward comprehensive security for autonomous systems.

The rapid adoption of AI agents across enterprise workflows has introduced new attack surfaces, as these systems interact with tools, access sensitive data, and execute multi step processes. Traditional security approaches, which focus primarily on prompt filtering or static testing, are no longer sufficient to manage these risks.

Lee Weiner, CEO of TrojAI, emphasized the need for deeper visibility into AI operations. “The innovations we are unveiling this week address some of the most significant and rapid changes to the AI security ecosystem. Enterprise deployment of agents is accelerating quickly, and these new TrojAI capabilities enable a new level of visibility and protection needed for the Agentic enterprise,” said Lee Weiner, CEO of TrojAI. “Enterprises need to understand exactly what their AI agents are doing and to enforce policy across entire workflows, not just prompts. This is fundamental to deploying AI securely at scale.”

One of the key additions is Agent Led AI Red Teaming, which uses coordinated autonomous agents to simulate attacks on AI models, applications, and systems. These agents can execute complex multi step attack scenarios, automatically correlate findings, and map results to established security frameworks such as OWASP, MITRE, and NIST. The system also incorporates adaptive learning, allowing testing strategies to evolve based on previous results.

Another major capability is Agent Runtime Intelligence, currently available in private preview. This feature provides detailed visibility into how AI agents behave during execution, capturing data on tool usage, memory interactions, data retrieval patterns, and system prompt exposure. By analyzing these execution traces, organizations can identify risks such as excessive permissions, prompt injection propagation, and unauthorized data access across workflows.

The platform also introduces real time protection for AI coding agents, addressing a growing risk as these tools become embedded in software development processes. The system monitors coding agents as they generate and modify code, detecting exposed credentials, preventing sensitive data leakage, and blocking malicious instructions embedded within external inputs. This ensures that development workflows remain secure without disrupting productivity.

These capabilities are integrated into TrojAI’s broader platform, which includes dashboards, governance tools, and integrations with security information and event management systems. This unified approach allows organizations to enforce policies consistently across AI systems while maintaining visibility into their behavior.

The expansion of TrojAI’s platform reflects a broader evolution in enterprise AI security. As organizations increasingly rely on intelligent systems to automate workflows and decision making, the need for continuous monitoring, testing, and enforcement across the entire AI lifecycle is becoming critical.

By combining agent led testing, runtime intelligence, and real time protection, TrojAI aims to provide enterprises with the tools needed to manage risk and deploy AI systems confidently. The announcement highlights the growing importance of comprehensive enterprise AI security strategies in enabling safe and scalable adoption of advanced technologies.