Skip to main content
Back
DemonstrationGoogle Cloud
9 min read

Unit 42's Zealot AI agent demonstration

Environment
Unit 42's test environment
When
not statedDisclosed Apr 23, 2026
Entry
The chain starts inside Google Cloud: Unit 42 deployed the agent in a VM in its own isolated test environment, preconfigured with intentional vulnerabilities; an SSRF flaw in the port 3000 web service of a VM in a peered VPC reached the instance metadata service, which returned the attached service account's access token.

Palo Alto Networks Unit 42 built Zealot, a multi-agent AI system, and ran it inside a VM in an isolated Google Cloud environment that it had preconfigured with intentional vulnerabilities, with one instruction: exfiltrate sensitive data from BigQuery. The agent found a peered VPC, used an SSRF flaw in a connected VM's web service to take the attached service account's token from the instance metadata service, exported a BigQuery table to a bucket it created and granted itself storage.objectAdmin to read the export. Unit 42 published the report on April 23, 2026; it does not give the run's date or duration, and it says that in several scenarios human operators had to break logic loops.

What the demonstration did

Research team
Unit 42 researchers
Internet
Instance metadataservice account token
Zealot VM
BigQuery + Cloud Storage
+ web app, port 3000
+ open SSH port
+ service account
Peered VPC · connected VM
Supervisor Agent
Application Security Agent
Infrastructure Agent
Cloud Security Agent
Google Cloud APIs
BigQuery · production
New Storage Bucket
IAM Policy

Replaying Unit 42's Zealot demonstration / published April 2026; the run date is not stated

  1. 1

    Reconnaissance

    Not dated · before Apr 23, 2026

    From its VM in the test environment, the agent scans the host and cloud network and finds a peered VPC with a connected VM, SSH and port 3000 open.

  2. 2

    Initial access

    Not dated · before Apr 23, 2026

    The agent uses an SSRF flaw in the port 3000 web service to reach the connected VM's metadata service, which returns the attached service account's token; the Google Cloud APIs accept it.

  3. 3

    Cloud enumeration

    Not dated · before Apr 23, 2026

    With the token the agent lists IAM permissions and BigQuery datasets; an access attempt on the dataset labeled production returns Access Denied.

  4. 4

    Escalation and exfiltration

    Not dated · before Apr 23, 2026

    The agent exports the table to a bucket it creates, grants itself storage.objectAdmin when it cannot read the bucket, and reads the export.

Impact

No victim: inside Unit 42's own test environment the agent exported a BigQuery table from the dataset labeled production to a storage bucket it created, granted itself storage.objectAdmin and read the export; Unit 42 reports several scenarios in which human operators had to break logic loops, so the system was not fully autonomous.

Outcome

What the run shows about your Google Cloud

The agent chains known misconfigurations with minimal guidance. The peering and the open ports are your network design, and an SSRF flaw like this one is in your application code. The metadata server's header check does not stop code on your VM that sends the header; in this run the server returned the service account's token. What you control is what that token can do: give the VM's service account only the roles it needs, deny it setIamPolicy with an IAM deny policy so it cannot grant itself a role, and bound what it reaches with a principal access boundary and a VPC Service Controls perimeter.

Would the controls have stopped it?

In hindsight, not as a whole. 3 controls would have refused a step outright, and 9 controls would have limited one. 2 steps have no cloud-provider control. The counts below are control assessments, not attack stages: several can address the same step. The same replay with the controls in place before the incident shows where each phase would have been refused or bounded.

3
Prevented
9
Contained
2
Not covered

How it could have been prevented

The same replay with the controls in place. Green marks where a step is refused, blue where it is bounded, and red where no control reaches it. Open a row to read the control, why it acts, and its caveat.

Research team
Unit 42 researchers
Internet
Instance metadataservice account token
Zealot VM
BigQuery + Cloud Storage
+ web app, port 3000
+ open SSH port
+ service account
Peered VPC · connected VM
Supervisor Agent
Application Security Agent
Infrastructure Agent
Cloud Security Agent
Google Cloud APIs
BigQuery · production
New Storage Bucket
IAM Policy

Replaying Unit 42's Zealot demonstration / published April 2026; the run date is not stated

1 · Reconnaissance

1 Contained1 Not covered3 on other clouds

2 · Initial access

3 Contained1 Not covered6 on other clouds

3 · Cloud enumeration

2 Contained3 on other clouds

4 · Escalation and exfiltration

3 Prevented3 Contained9 on other clouds

Key facts

  • Unit 42 deployed Zealot inside a VM in an isolated Google Cloud environment that it preconfigured with intentional vulnerabilities; the mission prompt named the goal, exfiltrating sensitive data from BigQuery, and gave no attack steps
  • Zealot was built on LangGraph: a supervisor agent delegated to an Infrastructure Agent, an Application Security Agent and a Cloud Security Agent, and only the supervisor saw the shared attack state
  • The agent found a peered VPC and a connected VM with SSH and port 3000 open; an SSRF flaw in the port 3000 web service reached the instance metadata service and returned the attached service account's access token
  • An access attempt on the BigQuery dataset labeled "production" returned Access Denied; the agent exported the table to a bucket it created, granted itself storage.objectAdmin when it could not read the bucket, and read the export
  • In scenarios Unit 42 does not tie to this run, the agent injected private SSH keys for persistence without being asked, and reached a second service account with broader permissions through VPC peering
  • In several scenarios, human operators had to break logic loops to prevent resource exhaustion; the report gives no run date, duration, run count or model names

Sources

Every fact in this breakdown comes from the public reporting below. The replay is a model of that reporting, not a reconstruction of the environment it describes.

  1. 1.
    Can AI Attack the Cloud? Lessons From Building an Autonomous Cloud Offensive Multi-Agent SystemPalo Alto Networks Unit 42 · Apr 23, 2026 · ResearchArchitecture, test environment, mission prompt, attack chain, privilege escalation, exfiltration, persistence scenario, limitations, human intervention, conclusions, recommendations, MITRE techniques
  2. 2.
    AI Can Autonomously Hack Cloud Systems With Minimal Oversight: ResearchersSecurityWeek · Apr 23, 2026 · NewsPublication date, coverage, architecture summary, persistence scenario, limitations, recommendations, context
  3. 3.
    What Zealot Reveals About AI's Cloud Offensive CapabilitiesSecurity Buzz · May 11, 2026 · NewsAttack chain summary, hedges on scope, independent comment, recommendations
  4. 4.
    Disrupting the first reported AI-orchestrated cyber espionage campaignAnthropic · Nov 14, 2025 · Vendor reportContext, autonomy figure, motivation for the research

The estate in the replay follows the researchers' description of their own test environment. Verdicts describe what a control does when it is in place before the run. They are not an assessment of any company's security program.

More cloud attack breakdowns

All breakdowns