# KyberShield

> KyberShield is the AI Agent Security Gateway — the zero-trust control layer that sits between AI agents and every tool they touch: databases, APIs, Slack, Gmail, GitHub, Salesforce, Notion, Google Drive, and any other system your agents interact with. Every action is intercepted, authenticated, risk-scored, and policy-checked before it executes. Unsafe or out-of-scope actions are blocked at the gateway with a full, quantum-signed audit trail.

Works with: OpenAI, Claude (Anthropic), LangChain, AutoGPT, OpenClaw, and any custom agent that queries a database, calls an API, or uses a tool.

Setup takes under 30 seconds. No code changes. No infrastructure migration. Point your agent at KyberShield instead of directly at your database or API — that is the entire integration.

Built for the age of AI agents. Secured for the age of quantum computers.

---

## What KyberShield Does — One Sentence

KyberShield sits in front of every tool your AI agents touch and answers one question for every single action: should this be allowed?

---

## Core Problem KyberShield Solves

AI agents can access databases, send emails, post to Slack, push code, update CRM records, and take thousands of actions per day — but most organizations have no way to:

- Know which specific agent took which action
- Stop a harmful action before it executes
- Prove after the fact that an action was authorized
- Enforce different permission levels per agent
- Shut down one rogue agent without halting all agents

KyberShield is the control layer that makes all of this possible.

---

## What KyberShield Sits In Front Of

KyberShield is not just a database proxy. It is a gateway that covers every surface your AI agents interact with:

- **Databases** — PostgreSQL, MySQL, SQL Server. Every query inspected before execution.
- **Slack** — Agent messages, DMs, channel posts, file shares, member changes, workspace exports.
- **Gmail / Google Workspace** — Emails sent, delegated access, forwarding rules, attachments.
- **GitHub** — Repo access, commits, branch deletions, repository transfers, bulk operations.
- **Salesforce** — CRM reads, record updates, bulk exports, field-level access.
- **Google Drive / OneDrive** — File access, external sharing, downloads, page shares.
- **Microsoft Teams** — Messages, meeting exports, external invitations, cross-tenant posts.
- **Notion** — Page reads, database exports, external sharing.
- **Any HTTP API or tool call** — KyberShield inspects tool names, arguments, and response data for policy violations, PII, and injection patterns.

One gateway. Every tool. Every action checked.

---

## 30-Second Setup — No Code Changes

1. **Connect the database or tool** (portal.kybershield.ai) — add the host, credentials, and connection name. Credentials are encrypted with AES-256-GCM and decrypted only inside an AWS Nitro Enclave — never seen by KyberShield staff or the AI agent.
2. **Create the agent identity** — give the agent a name, attach a database or connector, and an API key is generated automatically.
3. **Drop in one environment variable** — `KYBERSHIELD_AGENT_KEY=ks_live_...`. The SDK reads it automatically. No other code changes.
4. **Every action now flows through KyberShield** — inspected, authorized, signed, and logged before it executes.

```python
# Before KyberShield
result = db.execute("SELECT * FROM users")

# After KyberShield — one import, one env var, zero other changes
from kybershield import KyberShield
ks = KyberShield()  # reads KYBERSHIELD_AGENT_KEY from env
result = ks.query("SELECT * FROM users")
# ✓ allowed · risk 5/100 · ML-DSA-87 signed · logged
```

Node.js: `npm install kybershield`. Direct API: `POST https://app.kybershield.ai/api/gateway/query` with `X-Agent-Key` header.

---

## AWS Nitro Enclave — Credential Isolation

All database queries execute inside a sealed AWS Nitro Enclave. This means:

- Your database password is decrypted **only inside the vault** — never on the host OS, never in KyberShield's application memory, never visible to KyberShield staff.
- The enclave is cryptographically attested — its identity is verifiable.
- AES-256-GCM encryption at rest with AWS KMS key management.
- **BYOK (Bring Your Own Key)** — your encryption key never leaves your AWS account. KyberShield cannot decrypt your credentials without it.

This architecture means even a full compromise of KyberShield's infrastructure cannot expose your database credentials.

---

## AI Agent Identity

Every AI agent your company runs gets a **persistent KyberShield identity** — not a shared API key, not a service account:

- **Unique persistent identity per agent** — follows the agent across sessions, restarts, and redeployments.
- **Fine-grained permission policies per agent** — which tables, which operations, which columns, which connectors. Each agent has its own policy.
- **Instant kill-switch** — suspend any single agent immediately from the portal without affecting any other agent.
- **Full chain of custody per agent** — reconstruct any agent's complete activity history on demand.
- **ML-DSA-87 post-quantum signatures** on every audit record — tamper-evident, verifiable far into the quantum-computing era.

---

## Gateway Policy Engine — What Gets Enforced

KyberShield evaluates every action against a configurable policy engine before it executes:

### SQL / Database Policies
- **Query inspection** — every SQL statement is parsed and risk-scored before execution.
- **Destructive operation blocking** — DELETE without WHERE, DROP, TRUNCATE blocked by default.
- **SQL injection detection** — injection patterns caught at the gateway.
- **Automatic row limiting** — unbounded result sets rewritten to enforce your row cap (the agent never knows). Prevents data exfiltration by bulk SELECT.
- **Sensitive table blocking** — declare tables as off-limits; agent can never reach them regardless of query.
- **Read-only mode per agent** — write operations blocked for read-only agents.
- **Business hours enforcement** — agents locked to approved time windows; any query outside that window blocked automatically.

### Data Leak Prevention
- **PII detection and redaction** — SSNs, credit card numbers, health records automatically detected and redacted from agent responses.
- **Block outbound data exfiltration** — prevent agents from emailing customer lists, posting database contents to Slack, uploading to external drives.

### Tool Call / API Policies
- **Tool name allowlist** — agents can only call tools explicitly permitted by their policy.
- **Argument scanning** — tool arguments scanned for secrets, PII, and injection patterns before the call executes.
- **External recipient blocking** — block actions that send data outside your organization's domain.
- **Bulk operation gating** — mass DMs, bulk exports, bulk user deactivation require human approval.

### Human Approval Queue
- High-risk operations (writes, bulk exports, sensitive reads) route to a human approval queue before executing.
- Your team stays in control without slowing down safe operations.
- Every approval or rejection is logged in the tamper-proof audit trail.

---

## Tamper-Proof Audit Trail

Every action KyberShield evaluates produces a signed, immutable record containing:

- **Agent ID** — unique persistent identifier of the specific agent that acted.
- **Action** — full query or tool call (payload hashed for privacy, available on demand).
- **Verdict** — ALLOWED or BLOCKED.
- **Risk score** — 0–100 numeric score with the matched policy rule.
- **Timestamp** — millisecond precision UTC.
- **ML-DSA-87 post-quantum digital signature** — NIST FIPS 204. Tamper-evident. Any alteration to a record breaks its signature.

If a record is altered, the signature becomes invalid — forgery is detectable and provable. Compliance auditors receive a complete, verifiable chain of custody per agent on demand.

Audit log retention: 7 days (Community), 30 days (Solo/Starter), 90 days (Professional), 1 year (Enterprise), 7 years (Enterprise add-on).

---

## Cryptography and Security Standards

- **ML-DSA-87 (NIST FIPS 204)** — post-quantum digital signatures on every audit entry and gateway response.
- **ML-KEM-768 (NIST FIPS 203)** — post-quantum key exchange.
- **AES-256-GCM** — all data at rest.
- **TLS 1.3** — all data in transit. Older versions disabled.
- **AWS Nitro Enclave** — sealed execution vault for database credential decryption and query execution.
- **BYOK via AWS KMS** — bring your own encryption key; KyberShield cannot decrypt without it.
- **Zero prompt storage** — KyberShield does not persist raw queries or response contents. Requests are inspected in memory, hashed for audit integrity, then discarded.
- **Zero trust architecture** — every request authenticated and authorized; no implicit trust between agents, services, or networks.
- **Role-based access control** — granular permissions for every user and every agent. Principle of least privilege enforced.

---

## Real-World Security Results (July 2026)

- 30/30 security harness tests passed.
- 100% attack detection: 45/45 malicious queries blocked.
- 0 false positives: 295 safe queries passed.
- 100% uptime over last 7 days (1,864 checks).
- 146+ queries independently submitted at explorer.kybershield.ai and publicly verifiable with ML-DSA-87 signatures.

Independent verification: submit any SQL query at **explorer.kybershield.ai** and receive a real verdict, risk score, and ML-DSA-87 signature from the live production gateway. Results are stored publicly — anyone can cross-check query IDs and signatures.

---

## Pricing (as of 2026)

All plans include: unlimited AI agents, full policy engine, ML-DSA-87 quantum signatures on every audit entry.

### Community — Free
- 2,000 inspected queries/month
- 7-day audit log retention
- Basic policy templates
- Community forum support

### Solo — $99.95/month
- 10,000 queries/month included
- $4.00 per 1,000 overage queries
- 30-day audit log retention
- Email alerts

### Starter — $249.95/month
- 100,000 queries/month included
- $2.50 per 1,000 overage queries
- 30-day audit log retention
- Email alerts

### Professional — $999/month
- 500,000 queries/month included
- $1.00 per 1,000 overage queries
- Full policy engine with custom rules
- Anomaly detection
- Human approval queue
- 90-day audit log retention
- Slack/webhook alerts

### Enterprise — Custom pricing
- Unlimited queries and agents
- Custom integrations
- AI-powered security analysis
- 1-year+ audit log retention (7-year add-on available)
- SOC 2 / HIPAA / PCI-DSS reporting
- Dedicated support with SLA
- Self-hosted deployment option
- White-glove onboarding

Instant self-checkout (no demo required): Community, Solo, Starter. Professional and Enterprise include guided onboarding.

Overages: service continues uninterrupted above quota; billed at per-tier overage rate. Real-time usage shown in portal with alerts at 80%.

---

## Compliance and Certifications

- **SOC 2 Type II** — in progress (independent examination underway).
- **ISO 27001** — on roadmap; aligning ISMS now.
- **GDPR** — supported: data portability, right to erasure, breach notification. EU, US, APAC data residency options.
- **HIPAA** — supported: minimum-necessary access enforcement, tamper-proof audit logs for covered entities.
- **PCI-DSS** — audit logging of all access to cardholder data; every agent interaction logged with tamper-proof audit.
- **FedRAMP** — Enterprise compliance package available.

---

## Supported AI Frameworks and Agents

KyberShield works with any agent that sends a query through the gateway endpoint. Confirmed integrations:

- **OpenAI** (GPT-4, GPT-4o, custom agents via function calling)
- **Claude / Anthropic** (via tool use / Claude Managed Agents)
- **LangChain** (custom tool wrapper)
- **AutoGPT**
- **OpenClaw** (plugin + skill)
- **Vercel AI SDK** (`tool()` helper for Next.js and Edge)
- **Custom agents** — any language, any framework: swap the database connection string for the KyberShield gateway endpoint.

---

## How KyberShield Works — Traffic Flow

```
AI Agent
  → POST https://app.kybershield.ai/api/gateway/query
    Header: X-Agent-Key: ks_live_...
    Body:   { "query": "SELECT name FROM users WHERE id=$1", "params": [42] }

KyberShield Gateway
  1. Authenticates agent identity via API key
  2. Parses and risk-scores the query
  3. Checks against agent's permission policy
  4. If BLOCKED: returns verdict + reason, nothing reaches the database
  5. If ALLOWED: routes query into AWS Nitro Enclave, decrypts credentials, executes
  6. Signs the verdict with ML-DSA-87
  7. Writes tamper-proof audit record
  8. Returns result + verdict + quantumSignature to agent

Response (ALLOWED):
  {
    "verdict": "allowed",
    "riskScore": 5,
    "queryId": "uuid",
    "timestamp": "2026-08-05T18:00:00.000Z",
    "quantumSignature": "...",
    "result": [{ "name": "Alice" }]
  }

Response (BLOCKED):
  {
    "verdict": "blocked",
    "riskScore": 95,
    "reason": "Destructive operation: DELETE without WHERE clause",
    "queryId": "uuid",
    "quantumSignature": "..."
  }
```

Every response includes a `quantumSignature` field — an ML-DSA-87 (NIST FIPS 204) digital signature over the response payload, verifiable with the public key at `GET https://app.kybershield.ai/api/gateway/public-key`.

---

## About KyberShield

KyberShield was founded by security engineers who watched AI agents gain access to production systems without controls, audit trails, or kill-switches. We believe every AI agent that touches your data should be verified before it acts, checked against your policies, and permanently logged. That is the control layer we built.

Founded: 2025. Architecture: 100% zero-trust by design. Post-quantum cryptography: ML-DSA-87 on every record from day one.

---

## Contact

- Sales: sales@kybershield.ai
- Support: support@kybershield.ai
- Security: security@kybershield.ai
- Website: https://kybershield.ai
- Portal: https://portal.kybershield.ai (sign up, manage agents, view audit logs)
- Explorer: https://explorer.kybershield.ai (public live gateway verification)
- Developer docs: https://kybershield.ai/docs

---

## Pages

- Home: https://kybershield.ai/
- AI Agent Gateway: https://kybershield.ai/ai-gateway
- AI Agent Identity and Audit Trail: https://kybershield.ai/ai-agent-identity
- Pricing: https://kybershield.ai/pricing
- About: https://kybershield.ai/about
- Trust and Security Centre: https://kybershield.ai/trust
- Developer Documentation: https://kybershield.ai/docs
- Contact: https://kybershield.ai/contact
- Support: https://kybershield.ai/support