Send events to Google Security Operations (Google SecOps, SIEM)
PoisonZero sends its events to Google SecOps as UDM events. Setup takes about 5 minutes. You need no parser and no forwarder.
What you need
- An Enterprise plan.
- A Google SecOps instance and the Google Cloud project it belongs to.
- A Google Cloud account that may create roles and grant access in that project, and the
gcloudcommand line (Cloud Shell works too). - No parser and no forwarder.
Step 1: Enter the instance in the console
- Open Settings > Devices & Model and find the section Google SecOps. Turn on Enable export. As Method choose Chronicle API.
- Enter Project ID, Region and Instance ID. The region is the SecOps region ID, for example
euoreurope-west3. - Click Save. At the first save PoisonZero creates a service account for your tenant. Its e-mail address then appears under Service account.
Step 2: Grant access
Copy the e-mail under Service account and run the two commands. Replace PROJECT_ID and SA_EMAIL. The role only allows chronicle.events.import.
gcloud iam roles create poisonzeroSecopsImport --project=PROJECT_ID --title="PoisonZero SecOps import" --permissions=chronicle.events.import gcloud projects add-iam-policy-binding PROJECT_ID --member=serviceAccount:SA_EMAIL --role=projects/PROJECT_ID/roles/poisonzeroSecopsImport
No key and no secret changes hands. Remove the binding to withdraw the access at any time.
Step 3: Send a test event
Click Send test event. The console shows whether Google SecOps accepted the event.
Step 4: Check that events arrive
In Google SecOps, open the UDM search and run this search. It shows all PoisonZero events:
metadata.vendor_name = "PoisonZero"
The console shows the status under Last export: time, HTTP status, error and number of events. Events appear in SecOps within minutes.
What is sent
Every hit from exec0, poison0, leak0 and inject0 is sent as a UDM event with the vendor PoisonZero and the action BLOCK, QUARANTINE or ALLOW. Command text and file contents never leave PoisonZero. The details are in the sections below.
All events
| PoisonZero event | event_type | action | category | Severity |
|---|---|---|---|---|
exec0 blocks a tool calltoolgate_denied, toolgate_rule_override_denied | Depends on the tool, see below | BLOCK | POLICY_VIOLATION | From the threat level |
You deny a tool call that exec0 asked abouttoolgate_asked_denied | Depends on the tool, see below | CHALLENGE and BLOCK | POLICY_VIOLATION | From the threat level |
You allow a tool call that exec0 asked abouttoolgate_asked_allowed | Depends on the tool, see below | CHALLENGE and ALLOW | POLICY_VIOLATION | From the threat level |
A rule override lets a blocked call throughtoolgate_rule_overridden | Depends on the tool, see below | ALLOW | POLICY_VIOLATION | HIGH |
exec0 observed or did not check a tool calltoolgate_observed, toolgate_unchecked | Depends on the tool, see below | ALLOW | POLICY_VIOLATION | From the threat level |
poison0 quarantines or holds a changequarantine_reversible, hold_awaiting_review | SCAN_FILE | QUARANTINE | SOFTWARE_SUSPICIOUS | From the threat level |
poison0 keeps a changedetected_kept, hold_kept_review, observed_would_hold, observed_would_quarantine | SCAN_FILE | ALLOW (hold_kept_review: CHALLENGE and ALLOW) | SOFTWARE_SUSPICIOUS | From the threat level |
poison0 reverts a changerevert, meta_attack | SCAN_FILE | BLOCK | SOFTWARE_SUSPICIOUS | From the threat level |
leak0 blocks a leakleak0_blocked | GENERIC_EVENT | BLOCK | DATA_EXFILTRATION | From the threat level |
leak0 removes the secret from the outputleak0_redacted | GENERIC_EVENT | ALLOW_WITH_MODIFICATION | DATA_EXFILTRATION | From the threat level |
leak0 warns or observesleak0_warned, leak0_observed | GENERIC_EVENT | ALLOW | DATA_EXFILTRATION | From the threat level |
leak0 asksleak0_asked | GENERIC_EVENT | CHALLENGE | DATA_EXFILTRATION | From the threat level |
inject0 withholds fetched contentinject0_withheld | GENERIC_EVENT | BLOCK | SOCIAL_ENGINEERING | From the threat level |
inject0 marks fetched contentinject0_marked | GENERIC_EVENT | ALLOW_WITH_MODIFICATION | SOCIAL_ENGINEERING | From the threat level |
inject0 observesinject0_observed | GENERIC_EVENT | ALLOW | SOCIAL_ENGINEERING | From the threat level |
Bash, Shell and PowerShell are PROCESS_LAUNCH, all other tools GENERIC_EVENT. The severity follows the threat level. Use metadata.product_event_type and security_result.action in rules.
Fields and mapping
PoisonZero maps its audit events to UDM. Command text, file content and prompt text are not part of any event, they never leave PoisonZero. Sent are the tool name, the agent, the decision, the reason, the threat level, the name of the device and, for poison0, the path of the file.
Fields you can search and filter on:
metadata.vendor_nameandmetadata.product_name: always PoisonZero.metadata.product_event_type: the PoisonZero action from the first column.principal.hostname: the name of the device.security_result.action,security_result.severityandsecurity_result.category.security_result.rule_name: the reason, for example the rule that fired.
YARA-L sample rules
Paste each rule into the rules editor of Google SecOps. Run it as a test over the last few days first, then enable it.
exec0 blocked a tool call
rule poisonzero_exec0_block {
meta:
author = "PoisonZero"
description = "exec0 blocked a tool call of an AI agent"
severity = "HIGH"
events:
$e.metadata.vendor_name = "PoisonZero"
$e.metadata.product_event_type = /^toolgate_(asked_|rule_override_)?denied$/
$e.security_result.action = "BLOCK"
condition:
$e
}poison0 quarantined or held a change
rule poisonzero_poison0_quarantine {
meta:
author = "PoisonZero"
description = "poison0 quarantined or held a dangerous change to an agent file"
severity = "HIGH"
events:
$e.metadata.vendor_name = "PoisonZero"
$e.metadata.product_event_type = /^(quarantine_reversible|hold_awaiting_review)$/
$e.security_result.action = "QUARANTINE"
condition:
$e
}leak0 stopped an exfiltration
rule poisonzero_leak0_exfiltration {
meta:
author = "PoisonZero"
description = "leak0 stopped a secret from leaving a device"
severity = "HIGH"
events:
$e.metadata.vendor_name = "PoisonZero"
$e.metadata.product_event_type = "leak0_blocked"
$e.security_result.category = "DATA_EXFILTRATION"
$e.security_result.action = "BLOCK"
condition:
$e
}Alternative: webhook feed (OCSF)
If Domain Restricted Sharing blocks the access grant in step 2, PoisonZero sends the events in OCSF format to a webhook feed.
- In SecOps, create a feed under Settings > Feeds: source type Webhook, log type OCSF. Copy the endpoint URL and the secret key.
- In the Google Cloud console, create an API key restricted to the Chronicle API.
- In the PoisonZero console, set the method to Webhook feed (OCSF), enter URL, API key and secret, and send a test event.
Field values can differ from the UDM path.
Troubleshooting
| What you see | What to do |
|---|---|
HTTP 403 and Permission 'chronicle.events.import' denied | The role is missing or belongs to another account. Run the commands from step 2 again. PoisonZero retries after 5 minutes on its own, nothing is lost. |
service account still propagating | The new service account is not available everywhere yet. Wait a minute and click Send test event again. |
HTTP 404 or a message about an unknown instance | Project ID, region and instance ID do not belong together. Copy the three values from your SecOps instance, without spaces. |
SecOps not yet provisioned | The export is not set up for your environment yet. Contact PoisonZero support. |
| Step 2 is rejected because the Domain Restricted Sharing policy blocks the access grant | Use the webhook feed (OCSF) from the Alternative section. |
The webhook feed reports HTTP 401 or HTTP 403 | Check the API key and the secret. Create a lost secret again with Generate Secret Key. |
| The test works, but nothing shows up in SecOps | Events appear within minutes. Last export shows the status and the error. |
Read next
SIEM export (local files, private devices, setup without the console). CrowdStrike Falcon Next-Gen SIEM. OpenTelemetry export. Incidents and audit.