Zero-Trust AI Security • Auth0 Hackathon 2026

Project Fulcrum

The AI agent that doesn't hold your keys. Identity-less by default.

Secure cross-domain security auditing with Auth0 Token Vault, Fine-Grained Authorization (FGA), and CIBA human-in-the-loop approval for high-stakes actions.

View Audit Log

The Problem with Traditional AI Agents

Most AI agents hold admin keys in .env files with excessive agency. One prompt injection = catastrophic damage.

GITHUB_TOKEN=ghp_xxxx

Permanent admin tokens in .env files

agent.execute("delete all repos")

Agent executes without permission checks

// No audit trail

No visibility into what agent actually did

The Auth0 Security Trinity

Fulcrum uses three Auth0 services to create a zero-trust architecture where the agent never holds permanent credentials.

Token Vault

Your OAuth tokens (GitHub, Jira, Slack) are securely stored in Auth0's Token Vault. The agent only receives short-lived, scoped proxy tokens.

// Agent never sees this
GITHUB_TOKEN=ghp_xxx...

// Agent gets this instead
token = vault.getToken({
  scopes: ['repo:read'],
  ttl: '5m'
})

Credential isolation

Fine-Grained Authorization

Every action is validated against permission rules before execution. If the relationship doesn't exist, the agent cannot act.

// Before EVERY tool call
const allowed = await fga.check({
  user: 'user:alice',
  relation: 'can_execute',
  object: 'action:github_merge'
});

if (!allowed) throw DENIED;

Relationship-based access

CIBA (Human Approval)

High-stakes actions (delete, merge, modify) pause and send a push notification to your phone. Biometric approval required to proceed.

// Level 5 actions pause here
await ciba.initiate({
  userId: session.userId,
  bindingMessage: 
    "Fulcrum wants to delete
     branch: main"
});
// Agent waits for approval...

Human-in-the-loop

What Fulcrum Can Do

Secure multi-service orchestration across your developer tools

GitHub Integration

List and analyze repositories
Manage branches (Level 5)
Scan for exposed secrets
Create issues and PRs

Jira Integration

List projects and issues
Create security issues
Search with JQL queries
Update issue status

Slack Integration

List channels
Send security alerts
Notify on vulnerabilities
Coordinate with teams

Action Risk Levels

Every tool has an assigned risk level. Level 5 actions always require CIBA approval.

Level	Risk	Example Actions	CIBA Required
1	Read-only	List repos, get issue details, view projects	No
2	Safe scan	Scan for secrets, analyze code	No
3	Create	Create issues, send Slack messages	No
4	Modify	Update issues, create PRs	No
5	Destructive	Delete branch, merge PR, delete issues	YES

Ready to Try Zero-Trust AI?

Connect your services and let Fulcrum manage security auditing with full accountability.

Connect with Auth0 Manage Connections