Security Architecture

The safest place to deploy
autonomous AI agents.

Most AI platforms compete on features. We compete on trust. Rare Agents is built from the ground up with agent security, data ownership, and execution safety as first-class requirements — not afterthoughts.

AES-256-GCM Encryption
Solana Blockchain Anchoring
Prompt Injection Detection
TOTP Two-Factor Auth
Immutable Audit Logs
Real-Time ACE Scanning
Agent Kill Switch
Zero-Trust Architecture
Quantum-seeded Entropy

The honest truth about AI security

Most AI platforms today focus on speed, features, and UI. Very few are genuinely solving agent security, data ownership, and execution safety. Rare Agents is different because we face a risk that simple chatbots don't — autonomous agents that can take real actions on your behalf.

Agent Hijacking Risk

Attackers try to inject malicious instructions into agent memory and trick agents into exposing stored data. ACE blocks this at the layer level.

API Key Leakage Risk

Social integrations are a massive attack surface. Every token is encrypted at rest, scoped to minimum permissions, never stored in plain text.

Social Engineering Risk

Attackers don't hack systems — they hack people through agents. Every agent message is policy-scanned before it reaches you.

The Rare Agents Security Stack

Six layers of protection, purpose-built for autonomous AI agents.

Layer 1

Zero-Trust AI Architecture

Every request verified. Nothing assumed.

Every action — chat, tool call, vault access, API call — is authenticated and authorized independently. Agents cannot piggyback on each other's permissions and cannot escalate privileges across session boundaries.

  • JWT authentication on every API request
  • Agent-scoped permissions — no cross-agent data bleed
  • Stateless trust: each request stands alone
  • Tiered resource limits enforced server-side, not client-side
Layer 2

Agent Isolation Layer (ACE)

Your agents can't be turned against you.

The Agent Context Engine runs continuous health and policy scans across all agent activity. It detects prompt injection, phishing language, anomalous behavior, and policy violations — and either logs or hard-blocks them depending on your Security Mode.

  • Real-time prompt injection detection on every message
  • Standard Mode: flags and logs threats
  • Strict Mode: blocks and quarantines malicious inputs
  • System prompts are structurally locked in Layer 1 — user input cannot override them
  • 4-layer DIA hierarchy separates platform / tier / domain / agent instructions
Layer 3

Encrypted Personal Vault

Your data — nobody else can read it.

Every item stored in the Personal Data Vault is encrypted with AES-256-GCM using a key derived from your account. Data is encrypted before it leaves your session and only decrypted on-demand. Agents never receive raw vault content — only approved, scoped extracts.

  • AES-256-GCM encryption at rest for all vault items
  • Solana blockchain proof-of-existence for tamper detection
  • Agents cannot directly query the vault — they request, you approve
  • Server-side encryption at rest; authorized service features can decrypt content
  • OAuth connector tokens stored encrypted, never in plain text
Layer 4

Human + AI Verification Layer

Critical actions need human eyes.

Sensitive operations require explicit confirmation at multiple levels. TOTP 2FA protects account access, agent permission gates require in-session approval, and the social media workflow requires human review before any post goes live.

  • TOTP 2FA (per-user, opt-in)
  • Math captcha + honeypot bot prevention on registration
  • CrewAI social crew: human-in-the-loop approval before publishing
  • Vault items require explicit unlock before agent access
  • Plan Mode: agents propose before they execute
Layer 5

Real-Time Threat Detection

Attacks caught before damage is done.

The ACE engine runs automated scans every 5 minutes for health signals and every 15 minutes for policy compliance. Brute-force attempts, token exhaustion, abnormal posting rates, and injection patterns all trigger alerts surfaced in the admin panel.

  • Brute-force login detection with automatic flagging
  • Token exhaustion monitoring per-user per-session
  • Abnormal agent behavior triggers ACE health alert
  • Phishing language patterns detected in chat content
  • Prompt injection attempts logged with severity scoring
Layer 6

Trust Transparency Dashboard

Full visibility into what your agent did.

Every action your agent takes is logged to an immutable audit trail. The Trust Dashboard gives you a plain-language timeline of agent actions, active integrations, vault activity, and security events — so you always know exactly what is happening.

  • Immutable audit log for every agent action
  • Per-agent activity timeline viewable by the user
  • Active integration inventory with revoke controls
  • Vault access log: what was read, when, by which agent
  • Kill Switch: pause any agent instantly with one click

How we compare

What most platforms offer vs. what Rare Agents delivers.

Feature
Typical AI Platform
Rare Agents
Prompt injection protection
—
System prompt locking
—
Encrypted personal vault
—
Blockchain data anchoring
—
Runtime behavior monitoring
—
Human-in-the-loop agent approvals
—
Agent kill switch
—
Immutable per-agent audit log
Partial
TOTP 2FA
Partial

Deploy agents you can actually trust.

Join users who choose Rare Agents because they won't compromise on control, privacy, or safety.

Get Started Free

No credit card required · Cancel anytime