Pillar Security has launched SAIL 2.0, the latest version of its platform designed to secure AI agents throughout their lifecycle. The update focuses on runtime security and identity controls, areas that have become increasingly critical as organizations move beyond simple copilots and begin deploying autonomous AI agents with real permissions and system access.
As enterprises scale their use of agentic AI, many are discovering that traditional security tools are not built to handle the unique risks these systems create. AI agents can take actions across multiple systems, access sensitive data, and operate with varying levels of autonomy — often faster than existing controls can monitor or restrict.
SAIL 2.0 introduces new capabilities aimed at addressing these challenges, including improved visibility into agent behavior at runtime and stronger controls around agent identity and permissions. According to Pillar, the platform is designed to help security teams govern what agents can do, not just what they are prompted to do.
The release reflects a broader shift in the market. While much of the early focus in AI security centered on prompt injection and model-level risks, attention is now moving toward runtime controls and agent identity governance — two areas where current enterprise security stacks remain relatively immature.
Pillar Security positions SAIL 2.0 as a response to this gap, offering organizations a way to apply security policies directly to AI agents as they operate in production environments.
Conditions Driving This Change
Enterprises are rapidly moving AI agents from experimental pilots into production environments, where these systems are increasingly granted access to internal tools, data, and workflows.
Most existing security tools were built for human users or static applications and struggle to monitor or control the dynamic, autonomous behavior of AI agents that can take multi-step actions across systems.
The number of AI agents operating inside organizations is growing quickly, creating a large and often poorly governed population of non-human identities with varying levels of access and autonomy.
Traditional identity and access management systems lack the granularity needed to apply real-time, context-aware controls to AI agents that may act differently depending on the task or input they receive.
Runtime risks have become more visible, as several recent incidents have shown that prompt injection and malicious instructions can cause agents to perform unauthorized actions even when initial guardrails are in place.
Security and compliance teams are under increasing pressure to demonstrate control over AI systems, especially as regulators and auditors begin asking specific questions about how autonomous agents are governed in production.
Many organizations currently rely on a combination of manual reviews, basic logging, and overly broad permissions, which creates both operational friction and significant security gaps when dealing with agentic workflows.
The speed at which AI agents operate makes after-the-fact detection insufficient, pushing the market toward solutions that can enforce security and governance decisions at the moment of execution rather than after an action has already occurred.
What AI Agent Security Looked Like Before
Prior to the emergence of dedicated runtime security platforms, organizations had very limited options for securing AI agents once they moved beyond simple prompt-based interactions. Most companies relied on a combination of existing enterprise security tools that were originally built for human users or traditional applications. Identity and access management systems typically treated AI agents as service accounts, often assigning them broad, long-lived permissions across multiple systems with little ability to apply granular or context-aware restrictions.
Security teams generally depended on prompt-level guardrails and output validation to reduce risk. While these techniques helped block some obvious forms of prompt injection, they offered almost no protection once an agent began executing tasks autonomously across internal tools, databases, or APIs. In many cases, agents were granted wide-reaching access because there was no practical way to enforce narrow, task-specific boundaries in real time.
Monitoring and logging solutions existed but were largely reactive. Security and compliance teams could review logs after an agent had already completed a series of actions, but they had minimal visibility into what the agent was doing while it was operating. This created a significant blind spot, especially as organizations began deploying agents in more complex, multi-step workflows.
As adoption increased, the gap between the capabilities of AI agents and the controls available to govern them became more apparent. Many enterprises found themselves in a position where agents could access sensitive data and perform actions across different systems, often with permissions that exceeded what was strictly necessary for their intended function. Without runtime enforcement mechanisms, organizations were forced to either heavily restrict agent usage or accept elevated levels of risk. This dynamic slowed down adoption in higher-risk environments and created ongoing concerns for security, compliance, and audit teams.
What AI Agent Security Looks Like Now
The release of platforms like SAIL 2.0 reflects a shift toward applying security controls directly at the runtime layer of AI agents. Rather than relying primarily on input filtering or post-execution reviews, these solutions focus on monitoring and governing agent behavior while actions are being executed. This approach allows security teams to enforce policies based on what an agent is actually trying to do in a given context, rather than attempting to predict every possible input in advance.
SAIL 2.0 introduces more advanced identity and permission management capabilities specifically designed for autonomous systems. This includes the ability to apply dynamic, context-aware controls that can limit what an agent is permitted to access or execute based on the nature of the task. The platform also emphasizes improved visibility, giving organizations greater insight into how agents are operating across their environment in real time.
This evolution represents a meaningful change in how enterprises can approach AI agent governance. Instead of treating agents as static tools with fixed access rights, organizations now have the option to implement controls that adapt as agents perform different actions. Runtime enforcement reduces reliance on overly broad permissions and helps limit the potential blast radius if an agent receives unexpected or malicious instructions.
While these capabilities are still developing, they address several of the core challenges that previously made large-scale agent deployment difficult in regulated or high-risk environments. Security teams can now focus on defining acceptable behavior for agents and enforcing those boundaries during operation, rather than attempting to secure every possible interaction at the prompt level alone. This shift is helping organizations move from cautious experimentation toward more structured and governed use of agentic AI systems.
Our Take
AI Security Take
The launch of SAIL 2.0 highlights a clear shift in how organizations need to think about securing AI agents. While early efforts in AI security focused heavily on prompt injection and input filtering, it is becoming increasingly evident that these measures alone are not sufficient once agents are given real permissions and the ability to act autonomously across systems.
Runtime controls and agent identity governance are no longer optional considerations. As more companies move agents into production workflows, the ability to monitor, restrict, and audit what agents actually do while operating becomes a core requirement. Solutions that can enforce security decisions at the point of execution, rather than after the fact, will likely become a baseline expectation for any organization deploying agentic AI at scale.
For security and governance teams, this means shifting focus from solely protecting model inputs to building controls around agent behavior, permissions, and actions. Organizations that continue to rely primarily on broad access grants and reactive monitoring will face growing operational and compliance risks as their use of autonomous agents expands.
Pillar Security’s update reflects this broader industry movement toward runtime security for AI agents. While individual platforms will continue to evolve, the underlying need for stronger controls at the execution layer is unlikely to diminish. Companies evaluating agent security tools should prioritize solutions that offer granular, real-time visibility and the ability to enforce context-aware policies, rather than treating agent governance as an extension of traditional application security.
