
Console preview
Try the console
Try the console
without signing in
These are the screens our analysts work in every day, filled with example data.
Open a case to see why it got its verdict, then approve or reject the isolation request.
This preview runs on example data. None of it is real customer data, and the buttons don't trigger anything.
Open cases
214 alerts grouped into 3 cases (example)Suspected ransomware execution · CASE-2611-0412
- Evidence Email attachment invoice_1103.docm opened, then spawned a PowerShell child process (T1566.001, T1059.001)
- Evidence vssadmin.exe delete shadows /all /quiet executed (T1490)
- Evidence 412 file extensions changed within 6 seconds
- Reasoning All three behaviors happened on the same host within 7 minutes, in the same order as the execution stage of a ransomware attack
- Verdict Suspected ransomware execution · confidence 0.94
- Recommendation Isolate the host from the network (needs owner approval) · check the 4 mailboxes that received the same attachment
Record: input hash 9f3a…c21e · model v3.2 · policy P-07 · How to read a verdict record
Internal chatbot tried to send data out · CASE-2611-0381
- Evidence Hidden instruction inside an uploaded document ("Ignore previous instructions…")
- Evidence Chatbot called the customer lookup tool 12 times more often than usual
- Evidence Attempt to send to an external address with the mail.send tool (blocked by policy)
- Reasoning The tool calls followed the order given by the instruction in the document, which matches indirect prompt injection (OWASP LLM01)
- Verdict Tool misuse through prompt injection · confidence 0.88
- Recommendation End the session · block the document source · remove the chatbot's permission to send externally
Bulk download in the early hours · CASE-2611-0377
- Evidence VPN login at 02:14, outside the account's usual hours (09:00 to 19:00)
- Evidence 1,820 files downloaded from the design document repository in 40 minutes (usual daily average: 30)
- Evidence The account is on the list of upcoming departures (HR integration)
- Reasoning Far outside the account's baseline. This may be a handover period, so a person needs to check
- Verdict Possible insider data exfiltration · confidence 0.71
- Recommendation Restrict downloads for the account · confirm with the department head
Actions waiting for a person to approve
The AI never carries out high-impact actions on its ownIsolate FIN-PC-0412 from the networkCASE-2611-0412 · requested 09:42 · 6-step reasoning
Restrict downloads for account kim.***CASE-2611-0377 · requested 02:58 · confidence 0.71
Every approval or rejection is recorded with who made it, when, and which evidence they looked at.
Coverage map
Red team and blue team test results (example)Initial AccessTA0001
ExecutionTA0002
PersistenceTA0003
Privilege EscalationTA0004
Defense EvasionTA0005
Credential AccessTA0006
DiscoveryTA0007
Lateral MovementTA0008
CollectionTA0009
Command and ControlTA0011
ExfiltrationTA0010
ImpactTA0040
Select a cell to see the result for that technique.
October operations report
Example valuesAlerts received8,412Grouped into 61 cases
Cases where an analyst changed the verdict4Cause and rule fix attached
Verdicts issued without evidence0No evidence, no verdict
- Cases confirmed as false positives, their causes, and the detection rules we fixed
- Cases we missed and found later, and why we missed them
- Actions people approved or rejected
- Attack techniques to add to next month's red team testing
The metrics in this report and how we measure them are described in Measuring AI verdict quality.