Interoperable zero-trust permissions, cascading subagent delegation, semantic blast radius analysis, single-use replay defense, and cryptographically verifiable audit trails for AI agents.
This project and its intellectual property are available for commercial licensing or acquisition.
If you are an enterprise CISO, an AI platform looking to integrate an enterprise security layer, or an investor/acquirer:
📩 Open a Priority Inquiry: Click 💼 Submit Commercial Inquiry or connect via GitHub @SRN0129.
When companies deploy autonomous AI agents (LangChain, CrewAI, AutoGen, OpenAI Assistants, Anthropic Tool Use), they face a critical security problem: unrestricted AI side effects. If an agent hallucinates, enters an infinite loop, or suffers prompt injection, it can drain budgets, delete databases, or transfer company funds.
AgentTrust acts as the independent security gatekeeper:
Deny-by-Default Execution : No AI agent can execute tools or touch databases without explicit policy authorization.
2.
Cascading Subagent Delegation : Orchestrators can spawn worker subagents with strictly downscoped permissions (subagents cannot exceed parent capabilities).
3.
Semantic Blast Radius Scoring : Evaluates the real-world irreversibility of actions before execution. Actions with high blast radius dynamically escalate to human review.
4.
Human-in-the-Loop (HITL) Approvals with 1-Click Mobile Callback : Sensitive operations trigger approval tickets with 1-click decision URLs for Slack, Microsoft Teams, or mobile email.
5.
Single-Use Replay Protection : Approved tickets carry unique cryptographic nonces that burn atomically on first use. Duplicate attempts trigger security anomalies.
6.
Verifiable Audit Ledger & ECDSA P-256 Notarization : Every request and outcome is hash-chained (prev_hash$\rightarrow$ entry_hash ) and periodically signed with an asymmetric ECDSA P-256 keypair for offline compliance verification.
7.
Transparent Egress Proxy & Pre-Flight Prompt Scanner : Traps rogue outbound network calls and neutralizes prompt injections, jailbreaks, and sensitive data leaks.
8.
Emergency Air-Gap Killswitch : 1-click panic button that revokes all agent credentials in$< 15\text{ ms}$ and exports a forensic checkpoint.
| Threat Category | Attack Vector | Countermeasure |
|---|---|---|
| Spoofing | Attacker impersonates a privileged agent. | Scoped API keys (hashed with SHA-256 at rest); instant revocation; delegation lineage tracking. |
| Tampering | Rogue actor alters audit events to conceal unauthorized operations. | Hash chaining ( prev_hash + SHA-256 entry hash) + ECDSA P-256 signed checkpoints. |
| Repudiation | Operator denies authorizing autonomous transactions. | Complete audit trail capturing agent ID, operator email, payload hash, and approval records. |
| Information Disclosure | Secrets leaked via frontend code or plaintext logs. | API keys returned once at registration. Parameter payloads hashed at rest. |
| Elevation of Privilege | Subagent attempts to obtain permissions outside parent scope. | Strict capability attenuation boundary check rejects unauthorized delegation. |
| Denial of Service | Agent enters infinite loop, exhausting quotas. | Sliding-minute rate limiters (RPM) and cumulative budget ceilings ($ USD). |
| Replay Attacks | Attacker intercepts an approved human execution ticket. | Single-use nonces atomically burned ( CONSUMED ) on first use. Duplicate attempts return HTTP 409. |
| Air-Gap Containment | Widespread autonomous prompt injection compromise. | Global emergency killswitch revoking all credentials across all agents in |
AgentTrust features a rigorous test suite of 41 automated tests across 9 test suites covering all core security capabilities:
✓ tests/audit_integrity.test.ts (4 tests)
✓ tests/rate_limits_spend.test.ts (2 tests)
✓ tests/revocation.test.ts (4 tests)
✓ tests/authorization.test.ts (6 tests)
✓ tests/defensive_sandbox.test.ts (4 tests)
✓ tests/enterprise_defenses_v2.test.ts (7 tests)
✓ tests/approvals.test.ts (4 tests)
✓ tests/e2e_user_journey.test.ts (1 test)
✓ tests/enterprise_features.test.ts (9 tests)
Test Files 9 passed (9)
Tests 41 passed (41)
Run tests locally:
npm test
This repository includes pre-configured cloud deployment files (render.yaml and Dockerfile):
- Go to Render.com and sign in with GitHub.
- Click New +$\rightarrow$ Web Service$\rightarrow$ select
SRN0129/Agenttrust. - Select Free Plan and clickDeploy Web Service .
- Your platform will be online 24/7 with a live HTTPS URL.
| Endpoint | Method | Purpose |
|---|---|---|
/api/v1/agents/register |
POST |
Register root agent, generate scoped credentials |
/api/v1/agents/delegate |
POST |
Mint attenuated subagent with strictly downscoped capabilities |
/api/v1/authorize |
POST |
Core policy gateway check with semantic blast radius scoring |
/api/v1/proxy |
POST |
Transparent egress proxy inspecting and filtering outbound calls |
/api/v1/approvals/pending |
GET |
List pending human approval requests |
/api/v1/approvals/callback |
GET |
1-Click mobile push decision link for Slack, Teams, and email |
/api/v1/tools/execute |
POST |
Sandboxed protected tool execution gateway |
/api/v1/notarize/checkpoint |
POST |
Mint ECDSA P-256 signed audit checkpoint |
/api/v1/compliance/report |
GET |
EU AI Act (Art. 12 & 14) and NIST AI RMF compliance assessment |
/api/v1/system/killswitch |
POST |
Emergency air-gap killswitch revoking all agent credentials |
This software is released under dual-licensing terms:
- Open Source Evaluation : MIT License for personal and evaluation use.
- Commercial & Enterprise Use : For proprietary integration, dedicated support, or acquisition, contact@SRN0129 .