Veto

Product · Developer tool by Ona

Kernel-level enforcement for the next era of AI agents.

Veto is a kernel-level security engine that prevents AI agents from bypassing guardrails by enforcing policies at the syscall level. It uses content-addressable identification to block unauthorized actions in real time, regardless of file paths.

Launched
2026-03-03
Languages
English

Category: Cybersecurity

Deployment
Customer VPC (AWS, GCP, Azure)
Enforcement Layer
Kernel / BPF LSM
Identification Method
Content-addressable (Hash-based)
Enforcement Timing
Synchronous with syscalls
Deployment Model
Native to Ona platform in customer VPC
  • Kernel-level enforcement engine for AI agents
  • Content-addressable binary identification at the BPF LSM layer
  • Synchronous syscall-level enforcement with -EPERM returns
  • Prevents path-based denylist bypasses by reasoning agents

Overview

Veto is Ona's kernel-level enforcement engine designed specifically to secure autonomous AI agents. It addresses the unique challenge where reasoning-capable agents can bypass traditional path-based security tools by manipulating their own environment. By operating at the syscall level within the BPF LSM layer, Veto identifies and blocks unauthorized executables based on their content rather than their file path. This ensures that security enforcement remains below the agent's reach and is synchronous with execution. Veto is part of a defense-in-depth strategy for enterprises running agent fleets in high-consequence environments like banking and pharma.

Kernel-Level Security

  • Content-addressable identification: Veto operates at the kernel level to identify and block executables based on their content rather than just their file path.
  • Syscall-level enforcement: The system enforces security policies synchronously at the syscall level, returning -EPERM before a binary can execute.
  • Atomic check-and-enforcement: Veto uses the same kernel file reference for both the hash check and the policy enforcement to eliminate gaps between check and execution.
  • BPF LSM integration: The engine is built at the BPF LSM layer to provide deep visibility and control over agent actions.

Agent Governance

  • Absolute limit enforcement: Veto provides a hard outer boundary that defines absolute limits for repositories, tools, and system access.
  • Dynamic behavior boundaries_planned: The platform supports dynamic inner boundaries that adjust an agent's permissions based on real-time behavior.
  • Native runtime protection: Security policies are native to the runtime and kernel, preventing agents from manipulating the layer that enforces the rules.
  • Evasion prevention house: The system is designed to block agents from bypassing denylists by copying binaries to new paths or using /proc/self/root.

Use cases

  • Security teams use Veto to prevent AI agents from running unauthorized binaries even if they attempt to bypass path-based denylists.
  • Enterprises in regulated industries deploy Veto to ensure autonomous agents stay within defined operational boundaries at the kernel level.
  • Platform engineers implement Veto to provide a hard outer boundary for agent fleets operating across internal repositories and APIs.

Specifications

Deployment
Customer VPC (AWS, GCP, Azure)
Enforcement Layer
Kernel / BPF LSM
Identification Method
Content-addressable (Hash-based)
Enforcement Timing
Synchronous with syscalls
Deployment Model
Native to Ona platform in customer VPC

How it works

  1. Deployment in VPC: Veto is deployed as part of the Ona platform within the customer's own VPC on AWS, GCP, or Azure.
  2. Policy Configuration: Administrators define absolute limits for agent access to repositories, tools, and internal systems.
  3. Real-time Interception: When an agent attempts to execute a process, Veto performs a content-addressable hash check at the syscall level.
  4. Enforcement and Logging: The kernel enforces the policy synchronously, blocking unauthorized executions before they start.

FAQ

How does Veto prevent agents from bypassing path-based denylists?

Veto identifies executables by their content (hash) rather than their file path, preventing agents from bypassing security by moving or renaming binaries.

How can I get access to Veto?

Veto is currently available in early access for design partners with strict security requirements.

What types of actions can Veto block?

Veto enforces at the exec gate today, with plans to extend to network, file access, and memory primitives in the future.

Why is kernel-level enforcement necessary for AI agents?

Veto operates at the kernel level (Layer 3), which is below the agent's reach, preventing reasoning-capable agents from manipulating the enforcement layer.

Added by AntSpace from public sources (ona.com).

Related

More products by Ona