SynthSentinel

Continuous monitoring that detects and quarantines synthetic media in corporate communications.

Employees cannot trust every video call or voice note from leadership. SynthSentinel analyzes biometric consistency in real time and surfaces an instant green / amber / red trust flag before damage spreads.

Primary buyer	Enterprise security / communications integrity teams
Product	Browser & mobile SDK + operator console
Model	Per-seat subscription
Price	$15 / employee / month
Moat	Verified identity network effects

Why it matters

CEO and CFO impersonation fraud is already moving money. As synthetic voice and video become indistinguishable in meetings, Fortune 500 finance and banking teams need mandatory internal media verification — not another offline deepfake scanner.

Architecture

Client-side detection for privacy and low latency, with cloud escalation for hard cases and a policy engine that maps every verdict to an enforceable action.

flowchart TB
  subgraph Endpoints["Endpoints"]
    Zoom["Zoom / Teams / Slack"]
    Mobile["Mobile voice notes"]
    Ext["Browser extension"]
  end

  subgraph Edge["On-device / SDK"]
    Capture["Media capture"]
    Biometric["Biometric consistency<br/>analyzer"]
    Flag["GREEN · AMBER · RED<br/>UI overlay"]
  end

  subgraph ControlPlane["Control plane"]
    Gateway["API gateway"]
    Policy["Policy engine"]
    Core["Core orchestration"]
    Graph[("Policy graph")]
    Audit[("Immutable audit log")]
    Console["Operator console"]
  end

  Zoom --> Capture
  Mobile --> Capture
  Ext --> Capture
  Capture --> Biometric
  Biometric --> Flag
  Biometric --> Gateway
  Console --> Gateway
  Gateway --> Policy
  Gateway --> Core
  Policy --> Graph
  Policy -->|allow · flag · quarantine · block| Core
  Core --> Audit
  Core -->|escalate hard cases| Gateway
  Flag -.->|operator review| Console

Loading

Design principles

• Privacy-first — primary scoring runs on-device; raw media stays local when possible
• Bank-strict mode — uncertain verdicts quarantine by default
• Enforceable outcomes — every analysis resolves to allow, flag, quarantine, or block
• Auditability — decisions are written to an immutable activity trail for security review

Quick start

npm install
npm test
npm run demo

The demo flags an authentic Zoom-style call green and a synthetic Teams voice note red.

npm run console   # operator UI on http://localhost:4173

SDK usage

import { SynthSentinel } from "@aifrontiers/synthsentinel";

const sentinel = new SynthSentinel({
  orgId: "jpmorgan-comms-sec",
  strict: true,
});

const result = sentinel.analyze({
  streamId: "zoom-call-001",
  claimedIdentity: "ceo@jpmorgan.com",
  channel: "video",
  capturedAt: Date.now(),
  features: [/* on-device biometric vector */],
});

console.log(sentinel.flagLabel(result), result.action, result.reasons);

Repository layout

src/sdk/              Real-time media analysis client
src/policy/           Quarantine policy engine
src/audit/            Event / activity log
src/cli/              Local analysis CLI
apps/console/         Operator console
apps/demo-extension/  Chrome extension for Zoom / Teams
docs/                 Architecture notes

Roadmap focus

Ship the Chrome extension that overlays trust flags on live Zoom and Teams calls — see apps/demo-extension.

License

MIT