nagasatish chilakamarti

Posted on May 18

TealTiger v1.3.0 — Cryptographic Proof for Every AI Agent Decision

#ai #security #opensource #governance

AI agents are writing code, managing infrastructure, and making autonomous decisions. But who governs the governors?

We just released TealTiger v1.3.0 — deterministic, cryptographically verifiable governance for autonomous AI agents. Open source, Apache 2.0.

The Problem

Most AI governance falls into two traps:

"Trust but verify" — Log everything, review later. Damage already done.
"Block everything suspicious" — Use an LLM to judge another LLM. Non-deterministic, slow, expensive.

TealTiger takes a third path: deterministic enforcement with cryptographic proof. No LLM in the governance path. Same input + same policy = same decision, every time.

What's New in v1.3

🔐 Non-Human Identity (NHI)

Agents are principals, not ambient credentials:

Lifecycle management (active → suspended → revoked)
Scope enforcement — agents only access what they're allowed to
Zero Standing Privilege — just-in-time grants that expire
Attestation — verify agent integrity before granting access

🧾 TealProof — Cryptographic Receipts

Every governance decision produces a tamper-evident receipt:

SHA-256 decision hash chained via Merkle tree
RFC 3161 timestamping anchors
Standalone Verification SDK (no TealTiger dependency needed for audit)

⚡ TealFlow — Declarative Governance Workflows

name: high-risk-action-approval
on:
  agent_action:
    risk_score_above: 80
jobs:
  require-approval:
    steps:
      - uses: tealtiger/require-approval
        with:
          approvers: ['security-team']
          timeout: '30m'

Org-level workflows set a floor that teams cannot weaken.

🧠 TealClassifier — Local ML (≤20ms)

ONNX-based inference running locally. No API calls. Deterministic. Four ensemble modes to combine regex + ML detection.

📊 TealDrift — Behavioral Drift Detection

Statistical baselines per agent/provider/model. Alerts when behavior diverges beyond threshold.

💰 TealMonitor v2 — Governance-Owned Cost Ceilings

Application code cannot override governance limits. Anomaly detection for cost spikes. Reasoning-token budgets.

🛡️ OWASP Agentic Top 10 — Zero Config

const engine = new TealEngineV13({
  policy_packs: ['owasp-agentic-top10']
});
// All 10 ASI risks covered. One line.

🔌 Platform Adapters + 12 Providers

AWS Bedrock Agents, AWS AgentCore, Azure AI Agent Service
5 new providers: DeepSeek, Groq, Together AI, HuggingFace TGI, xAI
Total: 12 providers, 95%+ market coverage

Performance

Component	p50	p95	p99
TealEngine evaluate()	0.8ms	2.1ms	4.3ms
TealClassifier (ONNX)	12ms	18ms	22ms
TealGuard v2 (full)	1.5ms	3.2ms	5.1ms
TealProof hash	0.02ms	0.05ms	0.08ms

Install

npm install tealtiger@1.3.0
pip install tealtiger==1.3.0

Full Backward Compatibility

v1.2 configs work without modification. All new features are opt-in. No breaking changes.

DEV Community