Shopify says it has built an AI agent to search its production infrastructure for ways security safeguards might be bypassed. The notable twist is that the company describes the hunter as running against live systems, with a separate sandbox and several controls intended to keep its work contained.
Shopify Watch analysis
What happened
Shopify’s engineering blog describes Cyber Chaos Monkey as an agentic vulnerability hunter. Its Discovery component looks for potential weaknesses; a Sandbox component is used for testing. The company says the design also uses containerisation, egress proxies and a two-person integrity judge agent.
The stated goal is to find novel ways around safeguards before attackers do. Shopify’s account establishes what the company says it has built and how it describes the controls, not an independent assessment of the system’s effectiveness or safety.
Why it matters
AI agents are being given more room to take multi-step actions, and security teams are exploring whether they can use that capability to probe systems as well as operate them. Testing against production raises the stakes: the same ambition that might uncover a weakness also makes containment and oversight matter.
Shopify’s named controls give readers something concrete to examine beyond the usual promise that an AI system is “secure by design”. But the blog summary does not establish how often the agent finds useful flaws, what safeguards it has discovered, or how the company measures the risk of running it live.
Our read
This is a substantive example of AI being applied to defensive security, with a design that separates discovery from testing rather than treating an agent’s first idea as a green light. The interesting question is whether those controls work under pressure, not whether the agent has a dramatic name. Chaos, meet the change-control process.
What to watch
- Whether Shopify shares evidence of vulnerabilities found or fixes made.
- How the company evaluates the sandbox and the agent’s production access.
- Whether other organisations adopt similar agent-led security testing.
Discussion spark: Should security teams let AI agents probe live production systems when testing is isolated, or should that work stay in a separate test environment?
Sources and evidence
- Cyber Chaos Monkey: Staying ahead of security threats with AI agents (12 November 2026, 14:57 UTC)
Independent WittyWires tracker for public updates about Shopify. Not affiliated with or endorsed by Shopify; this is not an official account.