| Incident Type | Severity | Example | |----------------------|-----------|--------------------------------------------| | API key exposure | High

Responding to Security Incidents Involving OpenClaw

Q: Incident Response Checklist

Agent stopped All API keys rotated All platform tokens regenerated Logs exported for analysis Memory audited for sensitive data exposure Malicious skills removed Root cause identified Se

Q: Automated Incident Detection

yaml security: incidentdetection: enabled: true rules: - name: unusualactivity condition: "apicalls > 5x average" action: alert severity: medium - name: u

Clawpedia · For Humans

Incident response playbook for handling security breaches or vulnerabilities in your OpenClaw setup.

When Things Go Wrong

Security incidents involving AI agents require a specific response approach. Whether it is a compromised API key, a malicious skill, unauthorized access, or data exposure, this guide provides a structured incident response plan for OpenClaw.

---

Incident Types

Incident Type	Severity	Example

API key exposure	High	Key committed to public repository

Unauthorized access	High	Unknown user interacting with agent

Malicious skill	Critical	Skill exfiltrating data to external server

Data exposure	High	Personal data found in public logs

Prompt injection	Medium	Untrusted input manipulating agent behavior

Cost anomaly	Medium	Unexpected spike in API usage

Service compromise	Critical	Agent used to attack external services

---

Immediate Response Steps

Step 1: Contain


# Stop the agent immediately
openclaw stop --force

# If remote, kill the process
ssh server "kill -9 $(pgrep openclaw)"

Step 2: Assess


# Check recent activity
openclaw logs --since 24h --level warn
openclaw audit --since 24h

# Check for unauthorized access
openclaw logs --grep "unauthorized\|denied\|failed"

# Check running skills
openclaw skills list --running

Step 3: Revoke Credentials


# Rotate all API keys immediately
# OpenAI: https://platform.openai.com/api-keys
# Anthropic: https://console.anthropic.com/settings/keys

# Update config with new keys
export OPENCLAW_API_KEY="new-key-here"

# Revoke platform tokens
# Regenerate Telegram bot token via @BotFather
# Regenerate Discord bot token in developer portal
# Regenerate Slack tokens in app settings

Step 4: Investigate


# Export logs for analysis
openclaw logs --since 7d --json > incident_logs.json

# Export memory for review
openclaw memory export > incident_memory.json

# Check for unauthorized skills
openclaw skills list
openclaw skills audit --all

Step 5: Remediate

Based on the incident type:

Incident	Remediation

API key exposed	Rotate key, check usage logs at provider

Unauthorized access	Update access controls, add IP allowlisting

Malicious skill	Remove skill, scan for data exfiltration

Data exposure	Delete exposed data, notify affected parties

Prompt injection	Enable injection detection, sanitize inputs

Step 6: Resume


# Verify security configuration
openclaw security verify

# Start with enhanced logging
openclaw start --foreground --verbose

# Monitor closely for 24-48 hours
openclaw logs --follow --level warn

Cost anomaly	Set spending limits, review what triggered usage

---

Incident Response Checklist


[ ] Agent stopped
[ ] All API keys rotated
[ ] All platform tokens regenerated
[ ] Logs exported for analysis
[ ] Memory audited for sensitive data exposure
[ ] Malicious skills removed
[ ] Root cause identified
[ ] Security configuration hardened
[ ] Incident documented
[ ] Affected parties notified (if data exposure)
[ ] Agent restarted with enhanced monitoring
[ ] Post-incident review scheduled

---

Automated Incident Detection


security:
  incident_detection:
    enabled: true
    rules:
      - name: unusual_activity
        condition: "api_calls > 5x average"
        action: alert
        severity: medium
      - name: unauthorized_access
        condition: "auth_failures > 10 in 5m"
        action: lockdown
        severity: high
      - name: data_exfiltration
        condition: "outbound_data > 100MB in 1h"
        action: lockdown
        severity: critical
    notifications:
      - type: email
        to: security@company.com
      - type: slack
        channel: "#security-alerts"

---

Post-Incident Review

After resolving the incident, conduct a review:

Timeline — What happened and when?
Root cause — How did it happen?
Impact — What data or systems were affected?
Response — How effectively did we respond?
Prevention — What changes prevent recurrence?

Document findings and update security procedures.

---

Tips

Have a response plan ready before an incident occurs.
Practice incident response with simulated scenarios.
Rotate credentials regularly to limit exposure windows.
Enable automated detection to catch incidents early.
Keep offline backups of critical configurations.
Know your provider contacts — OpenAI, Anthropic, etc. have abuse reporting channels.

---

Troubleshooting

Problem	Solution

Cannot stop agent remotely	SSH and kill process manually

Logs are missing or corrupted	Check log rotation settings and disk space

Cannot rotate API key	Contact provider support directly

Incident keeps recurring	Root cause not properly identified; dig deeper

Unsure if data was exposed	Assume worst case and notify affected parties