# What CAIN Does NOT Claim

**Version:** 2.0
**Date:** 2026-09-01

---

## Honest Limitations

We maintain explicit limitations to increase credibility. If a claim is not listed here, CAIN does not make it.

---

## Security Guarantees CAIN Does NOT Make

### CAIN does not guarantee model correctness

CAIN enforces policy, but does not verify that the AI model itself is correct, unbiased, or safe. A correct enforcement of a flawed policy is still a flawed outcome.

### CAIN does not eliminate all security risk

CAIN reduces risk through enforcement and evidence, but cannot eliminate it. Security risk management requires multiple layers.

### CAIN does not automatically intercept arbitrary OS activity

CAIN only controls actions explicitly routed through CAIN. It cannot intercept actions that do not pass through the enforcement layer.

### CAIN does not certify organizations

CAIN does not certify that an organization is trustworthy, secure, or compliant. CAIN only provides enforcement and evidence for actions within CAIN.

---

## Validation Claims CAIN Does NOT Make

### CAIN does not claim independent validation without an independent validator

If CAIN claims something is "independently verified," there must be an actual independent verifier with public source code and reproducible results.

### CAIN does not claim certification without a real assessor

CAIN does not claim certifications (SOC2, HIPAA, etc.) unless conducted by actual certified assessors with published reports.

### CAIN does not claim penetration testing without results

If CAIN claims penetration testing, the results must be published with scope, methodology, and findings.

---

## Performance Claims CAIN Does NOT Make

### CAIN does not guarantee specific latency

Latency depends on network, load, and implementation. CAIN measures and publishes actual latency, not guaranteed latency.

### CAIN does not guarantee 100% availability

No system is 100% available. CAIN publishes measured availability and recovery capabilities.

### CAIN does not guarantee throughput under all conditions

Throughput depends on load, network, and configuration. CAIN publishes measured throughput.

---

## Implementation Claims CAIN Does NOT Make

### CAIN does not claim implementations are bug-free

All software has bugs. CAIN publishes known limitations and issue trackers.

### CAIN does not claim perfect forward secrecy

Security properties depend on implementation details. CAIN documents actual security properties.

### CAIN does not claim immunity to zero-day vulnerabilities

No system is immune. CAIN maintains security response procedures.

---

## Evidence Claims CAIN Does NOT Make

### CAIN does not claim evidence is tamper-proof

Evidence is immutable and verifiable, but not physically tamper-proof. Evidence integrity can be verified.

### CAIN does not claim evidence cannot be deleted

Evidence can be deleted by operators. CAIN provides evidence integrity verification for the evidence that exists.

### CAIN does not claim evidence is legally admissible

Legal admissibility depends on jurisdiction and context. CAIN provides verifiable evidence, not legal advice.

---

## Protocol Claims CAIN Does NOT Make

### CAIN does not claim protocol completeness

The protocol may not cover all use cases. CAIN publishes known gaps.

### CAIN does not claim backward compatibility forever

Breaking changes will be versioned and documented.

### CAIN does not claim protocol stability without versioning

Protocol changes follow semantic versioning.

---

## Operational Claims CAIN Does NOT Make

### CAIN does not claim data residency by default

Data residency depends on deployment configuration.

### CAIN does not claim automatic backup verification

Backups must be verified by operators.

### CAIN does not claim incident response without procedures

Incident response requires defined procedures beyond CAIN.

---

## What CAIN DOES Claim

| Claim | Evidence |
|-------|----------|
| Enforcement works as documented | Public proof artifacts |
| Evidence is verifiable | Independent verifier |
| Decisions are logged | Evidence bundles |
| Fail-closed is implemented | Security invariant tests |
| Protocol is published | OpenAPI, schemas |
| Independent verification is possible | Two verifiers exist |

---

## The Principle

> "Trust is built through honesty about limitations."

CAIN's credibility comes from being honest about what it cannot do, not from claiming it can do everything.

---

*This document is updated with each release. Last updated: 2026-09-01*