EDR alert triage
How an AI SOC triages CrowdStrike and SentinelOne alerts: rebuild the process tree, check files and the user, and stage host isolation for approval.
The short answer
EDR alert triage decides whether an endpoint alert is malware, hands-on-keyboard activity, or expected developer tooling. Cotool's example Endpoint Alert Responder treats Falcon alert and behavior data as the source of truth and reconstructs the full process tree before judging any single event.[1]
The problem
Many endpoint alerts start with a user opening a malicious file, such as a document, archive, or executable delivered by phishing.[6]
A single event rarely shows intent. Judging it takes the parent processes, the user's sessions, and what the host did next, which means working across endpoint, identity, and cloud data.[1]
How an AI SOC handles it
Tune benign tooling
For benign developer tooling, the agent proposes a scoped exclusion instead of a broad suppression.[1]
Example
from falconpy import Hosts
falcon = Hosts(client_id=CLIENT_ID, client_secret=CLIENT_SECRET)
details = falcon.get_device_details(ids=[device_id])
# Only after a person approves containment
falcon.perform_action(action_name="contain", ids=[device_id], note="Approved containment")The two FalconPy calls behind the host steps: read the device, then contain it once a person approves. Containment needs an API client with Hosts: WRITE.[7]
What people approve
How Cotool handles it
- Cotool's response agents can be triggered by CrowdStrike Falcon detections, and the example Endpoint Alert Responder's tools include querying SIEM logs, revoking Okta user sessions, and taking action on hosts.[1]
- Agents hand work to investigator and reviewer subagents, or call other agents as tools, for parallel investigations and peer review.[4]
Integrations used
FAQ
- Does the agent isolate hosts on its own?
- Only if you let it. Isolating a host is a state-changing action that can require approval in the app or Slack, and one click makes an agent read-only.[4]
Sources
- Respond to every signal (Cotool Respond) · accessed November 16, 2026
- Triggering agents (Cotool docs) · accessed November 16, 2026
- Planning modes (Cotool docs) · accessed November 16, 2026
- AI SOC (Cotool) · accessed November 16, 2026
- Alerts (Cotool docs) · accessed November 16, 2026
- User Execution: Malicious File, T1204.002 (MITRE ATT&CK) · accessed November 16, 2026
- Hosts (FalconPy) · accessed November 16, 2026
- Threats (SentinelOne management API reference) · accessed November 16, 2026
- VirusTotal API v3 overview (VirusTotal docs) · accessed November 16, 2026



