Fresh Branch Reruns
OpenAI 89.1% · Google 85.5%
Current hardened-branch attack-behavior reruns across the canonical 55-payload slice.
Observe-Only Detection
OpenAI 20.0% · Google 72.7%
Fresh current-branch detection rates with L1/L2 still deterministic at 100% and 0.0% false positives in these reruns.
Verification Baseline
906 TS tests · 138 Python tests
Release baseline verified locally across Core, Gateway paths, and the Python SDK package surface.
Performance + Evidence
p50 52μs · Hist. N=525
Performance figure remains the historical benchmark; fresh reruns are shown separately so this demo does not blend evidence sets.
Choose an Attack Scenario
Select a scenario to begin.
Run The Proof
1. Run the clean control. 2. Run the protected attack. 3. Install Core and wrap the guarded outbound tool.
3. Install Core
Two-Line Integration
The demo ends in the same place the product does: install Core and wrap the guarded outbound tool.
Full Quickstart
npm install @cerberus-ai/core import { guard } from '@cerberus-ai/core'; const { executors: tools } = guard(rawTools, config, ['sendEmail']);
Start with 1. Run Control, then 2. Run Protected Attack
Watch the same runtime path move from normal behavior to a blocked outbound attack, then install Core from the docs.
What Cerberus Is Watching
Private Data Access
Reading sensitive files or records
Injection Detected
Hidden instructions in external content
Outbound Attempt
Trying to send data outside the org
Memory Poisoning
Malicious content stored for later
Threat Level
0/4
Cerberus Decision WAITING
Incident Intelligence Metrics Dashboard ↗
Select a scenario to see what's at stake.
Demo Progress Start with 1. Run Control.
1
1. Run Control Show a normal trusted-data workflow that completes without a block.
2
2. Run Protected Attack Replay the lethal runtime path and watch Cerberus interrupt the outbound action.
Ready 0 runs
Enterprise Scale
One Day at a Fortune 100
127,000 employees · 23 countries · 8 AI-powered departments · What Cerberus monitors in 24 hours
0
Agent actions monitored
0
Threats stopped
$0
Potential losses prevented
0.0%
Fresh rerun false positives
PeopleAssist AI
Human Resources
0
employee sessions
FinanceAssist AI
Finance & Payroll
0
transactions reviewed
ITAssist AI
IT & Infrastructure
0
systems monitored
CXAssist AI
Customer Success
0
account interactions
ExecAssist AI
Executive Operations
0
pipeline reviews
DevAssist AI
DevOps / CI-CD
0
pipeline runs
SecAssist AI
Security Operations
0
threat events scanned
Live Threat Feed
Monitoring… no threats detected yet.