Sysdig AI Defense
by Sysdig
Runtime security and governance that inventories AI agents and blocks risky actions before they complete
Sysdig AI Defense is a runtime security and governance product for security teams at companies running AI agents such as Claude Code, Codex, GitHub Copilot and Cursor. It tracks what agents actually do from laptop to cloud, ties each agent to an accountable person, and stops actions that break policy.
Sysdig AI Defense reached general availability on November 4, 2026, for macOS and Windows. It comes from Sysdig, the company founded by Loris Degioanni, creator of the Falco runtime security project. The product pairs instrumentation inside an agent's harness, which records tool calls, MCP servers, skills and plugins, with kernel-level operating system telemetry that records the processes those calls produce. A live action graph compares what the agent reported with what it did, and Sysdig treats a disagreement between the two as a detection. Sysdig reports roughly seven underlying processes for each action in agent logs, which Unite.ai describes as an early company finding. Policies start in alert mode and can move to enforcement, where they block unvetted MCP servers, skills and plugins or hold risky credential reads for human review. Templates map to the OWASP Top 10 for LLM Applications. Session records show which policy applied and who approved exceptions, and the product attributes model calls to agents, teams and people, including calls that bypass an AI gateway. The release also describes coverage of production Linux, Kubernetes and managed agent platforms, though Unite.ai notes that their availability status is unconfirmed and that no public matrix shows which controls prevent and which only detect. Sysdig leaves model scanning and red teaming to other tools. Bastion is the only customer named in the launch release, and a 14-day free trial is available.
A CISO or security engineering lead whose developers already use coding agents on company laptops and who has no record of what those agents did.
An agent inventory tied to named owners, with risky actions held or blocked and an audit record the agent cannot edit.
At a Glance
- Category
- Governance & Security
- Pricing
- Contact for pricing
- Target Market
- CISOs, Security Engineering Teams, CIOs, Enterprise Developers
- Deployment
- Cloud-first
- Founded
- 2013
- Headquarters
- San Francisco, United States
Key Features
- ✓Live agent inventory
Builds an inventory of agents, harnesses, models, MCP servers, skills and plugins from runtime telemetry, and links each agent to a responsible person.
- ✓Live action graph
Compares what an agent reported with what kernel telemetry shows it did, so a compromised or drifting agent stands out.
- ✓Policy enforcement
Policies start in alert mode, then block unvetted tools or hold risky credential reads for human review once the team is ready.
- ✓Audit records
Keeps session records of which policies applied and who approved exceptions, which gives auditors evidence the agent itself cannot rewrite.
- ✓AI spend attribution
Ties every model call to a person and a team, including calls outside the AI gateway, which helps finance and platform owners track costs.
Use Cases
- •Coding agent governance
A security team discovers which agents run on developer laptops and assigns an owner to each, which closes the shadow AI gap.
- •Credential protection
A policy holds an agent's risky credential read for human review, so a person decides before the action can complete.
- •Incident investigation
Investigators replay an agent session and compare what it intended, what ran and what changed, which speeds up finding the cause.
- •Compliance evidence
When an auditor asks what an agent did, the team produces records showing the applied policy and the person who approved it.
Ideal For
Best For
- ✓Governing coding agents such as Claude Code, Codex, GitHub Copilot and Cursor on employee macOS and Windows machines
- ✓Finding and blocking unvetted MCP servers, skills and plugins
- ✓Producing audit records of which policy applied to an agent action and who approved it
- ✓Attributing model spend to teams and people, including calls that bypass an AI gateway
Not Ideal For
- ✗Teams that need model scanning or AI red teaming, which Sysdig says this product does not do
- ✗Buyers who need confirmed Linux, Kubernetes or managed platform enforcement today, who should get that coverage in writing, since the general availability statement names macOS and Windows
- ✗Teams wanting a published price, since Sysdig offers only a trial, a demo and quotes
Deployment
Market Analysis
Pros
- ✓Observes agents below the application layer, so the record does not depend on what the agent reports
- ✓Alert mode lets a team tune policies before turning on enforcement
- ✓A 14-day free trial is open to companies that are not yet Sysdig customers
Cons
- ✗General availability is stated for macOS and Windows only, and Unite.ai could not confirm the status of Linux, Kubernetes and managed platforms
- ✗No public matrix shows which controls prevent an operation and which only detect it, across operating systems and harnesses
- ✗The product is new, with one named customer and vendor-sourced capability claims; SecurityWeek cites no independent evaluation
- ✗Toolradar reports the wider Sysdig platform as costly under per-node pricing, with a learning curve
Pricing
AI Defense
Contact for pricing
- ✓14-day free trial
- ✓Agent inventory
- ✓Policy enforcement
- ✓Session records and audit logs
Sysdig lists no price for AI Defense; its pricing page links to a trial and asks buyers to request a quote. Other Sysdig security products are licensed by host count, but the metering for AI Defense is not stated. Existing customers go through their account teams.
Security & Compliance
Sources
This page was written from 7 sources, 5 on domains other than sysdig.com.
- 1.sysdig.com — ai defensevendor
- 2.sysdig.com — pricingvendor
- 3.finance.yahoo.com — sysdig launches sysdig ai defense 150000056
- 4.unite.ai — sysdig ai defense agent runtime security
- 5.securityweek.com — sysdig adds runtime security and governance for ai agents
- 6.toolradar.com — sysdig
- 7.getlatka.com — funding
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Sentra
AI data security platform that maps which people and agents can reach sensitive data and removes excess access
MuleSoft Agent Fabric
A control plane to register, route, govern and cost-cap AI agents and MCP tools from any vendor
Goodfire
Interpretability platform for debugging AI models and guardrailing agents with low-cost activation probes
Asymptote Beacon
Open-source endpoint telemetry for AI coding agents, with a paid platform for agent policy and governance