Security Policy

1. Overview & Scope

This Security Policy describes the administrative, technical, and physical safeguards Karone Cognitive Enterprise Inc. maintains to protect the confidentiality, integrity, and availability of the Platform and all customer data processed therein.

This policy applies to:

• All Karone systems, networks, and infrastructure used to deliver the Platform
• All Karone employees, contractors, and third-party vendors with access to Platform systems
• All customer data processed, stored, or transmitted via the Platform
• All AI agent operations, including data inputs, model inference, and output handling

2. Security Governance

2.1 Security Leadership

2.2 Security Program

Karone maintains a formal Information Security Management System (ISMS) covering:

• Risk Assessment: Annual enterprise risk assessments and continuous threat modelling
• Policy Reviews: All security policies reviewed quarterly
• Security Roadmap: Published internally with quarterly milestones
• Executive Reporting: Monthly security metrics reported to executive leadership
• Board Oversight: Quarterly security briefings to the Board of Directors

2.3 Risk Management

3. Data Classification

AI agents handle Confidential data or above. All AI agent data pipelines enforce the controls of the highest-classified data type they process.

4. Access Control & Identity Management

4.1 Principle of Least Privilege

All access to Platform systems and customer data is granted on a least-privilege basis. Access is reviewed quarterly and revoked immediately upon role change or termination.

4.2 Authentication Requirements

4.3 Role-Based Access Control (RBAC)

The Platform implements granular RBAC with predefined and custom roles. Tenant Administrators can define roles with specific permissions across modules and restrict access by user, department, cost centre, and approval tier.

4.4 Privileged Access Management

• All privileged access to production infrastructure requires approval via a PAM system
• Privileged sessions are recorded and retained for 12 months
• No standing privileged access — all elevated access is time-bound (just-in-time)
• Break-glass accounts are sealed, monitored, and require dual authorization

4.5 Session Management

• Sessions expire after 30 minutes of inactivity (configurable by Tenant Administrator)
• Concurrent session limits enforced per user
• Session tokens are cryptographically signed and rotated on privilege escalation
• Forced re-authentication required for sensitive operations (e.g., changing payment details, exporting bulk data)

5. Network Security

• Web Application Firewall: WAF deployed at the edge blocking OWASP Top 10 attacks, SQL injection, XSS, and rate-limit violations in real time
• DDoS Protection: Multi-layer DDoS mitigation via cloud provider shield + Cloudflare, with automatic traffic scrubbing
• Network Segmentation: Production, staging, and development environments are fully isolated via separate VPCs and network ACLs
• Zero Trust Architecture: All internal service-to-service communication is authenticated and encrypted — no implicit trust based on network location
• TLS 1.3 Everywhere: All data in transit encrypted with TLS 1.3; TLS 1.0 and 1.1 disabled; certificates managed via Let's Encrypt
• Intrusion Detection: Network IDS/IPS deployed across all environments with real-time alerting for anomalous traffic

5.1 Egress Controls

Outbound connections from Platform services are restricted to an approved allowlist. All outbound traffic is logged and monitored for data exfiltration patterns.

5.2 API Gateway Security

• All API traffic routes through a centralized API gateway enforcing authentication, rate limiting, and request validation
• Rate limits: 1,000 requests/minute per API key (configurable per plan)
• API keys are scoped by permission and expire automatically if unused for 90 days

6. Application Security

6.1 Secure Development Lifecycle (SDL)

Design (threat modelling) → Development (secure coding standards, no secrets in code) → Code Review (security-focused peer review) → Testing (SAST/DAST on every build) → Deployment (immutable infrastructure, CI/CD security gates) → Post-Deploy (continuous monitoring, periodic penetration testing)

6.2 OWASP Top 10 Controls

6.3 Dependency Management

All third-party dependencies are scanned for known CVEs on every build. Critical vulnerabilities patched within 24 hours; high within 7 days.

7. AI & LLM Security

7.1 Prompt Injection Prevention

Prompt injection attacks attempt to manipulate AI agents by embedding malicious instructions in user-supplied data. Karone defends against this via:

• Strict separation of system instructions and user-supplied content in all LLM prompts
• Input sanitization and validation before data enters AI pipelines
• Output validation to detect and reject anomalous or policy-violating AI responses
• Canary tokens in system prompts to detect exfiltration attempts
• Human-in-the-loop checkpoints for high-risk AI operations

7.2 Tenant Isolation for AI Agents

• Each tenant's AI agent context is fully isolated — no cross-tenant data leakage is architecturally possible
• AI agents operate within scoped permission boundaries and cannot access data outside their configured scope
• Tenant data is never used to fine-tune or train AI models for other tenants
• AI agent memory and context windows are cleared between sessions by default

7.3 LLM Provider Security

7.4 AI Output Validation

All AI agent outputs are validated before being acted upon, checking for: policy violations, anomalous financial values outside approval thresholds, hallucination indicators, and harmful content.

7.5 Model Security & Integrity

• AI model weights stored in isolated, access-controlled repositories
• All model updates tested in a sandbox before production deployment
• Model versioning maintained; prior versions can be rolled back within 1 hour
• Adversarial robustness testing performed on all models before deployment

8. Cloud & Infrastructure Security

8.1 Cloud Provider Security

8.2 Infrastructure Hardening

• Immutable infrastructure: Servers are never patched in place — new images are built and deployed
• CIS Benchmarks: All OS and container images hardened to CIS Level 2 benchmarks
• Container security: Kubernetes pods run as non-root; images scanned before deployment
• Secrets management: All secrets in HashiCorp Vault or cloud-native secret stores; never in environment variables
• Patch management: Automated patching; critical patches within 24 hours

8.3 Tenant Data Isolation

All customer data is logically isolated at the application and database layer using tenant-scoped identifiers. Cross-tenant data access is architecturally impossible through normal application paths. Dedicated infrastructure available on Enterprise plans.

8.4 Availability & SLA

9. Encryption & Key Management

9.1 Encryption Standards

9.2 Key Management

• Encryption keys managed via Azure Key Vault / AWS KMS
• Master keys (KEKs) are Hardware Security Module (HSM)-backed
• Data Encryption Keys (DEKs) rotated automatically every 12 months
• Key access is logged, audited, and restricted to authorized service identities
• Customer-Managed Keys (CMK) available on Enterprise plans

10. Vulnerability Management

10.1 Scanning Program

• Continuous: Automated SAST/DAST on every commit and pull request
• Daily: Container image scanning and dependency CVE checks
• Weekly: Authenticated infrastructure vulnerability scans
• Quarterly: External penetration test by an approved third-party firm
• Annually: Full red team exercise and social engineering assessment

11. Incident Response

11.1 Incident Classification

11.2 Response Process

• Step 1 — Detect: SIEM alerts, user reports, or automated anomaly detection triggers an incident ticket
• Step 2 — Triage (within 15 minutes): On-call security engineer assesses severity, assigns classification, and begins incident log
• Step 3 — Contain (within 1 hour for P1): Isolate affected systems, revoke compromised credentials, block malicious IPs, preserve forensic evidence
• Step 4 — Notify: Affected tenants notified within 72 hours of confirmed breach; regulatory notifications as required by GDPR / PIPEDA
• Step 5 — Eradicate & Recover: Root cause identified and eliminated; systems restored from clean backups; enhanced monitoring activated
• Step 6 — Post-Mortem (within 5 business days): Written post-mortem published internally; customer summary provided on request

11.3 Breach Notification

In the event of a confirmed personal data breach, Karone will notify affected Tenant Administrators within 72 hours of confirmation, in accordance with GDPR Article 33 and PIPEDA breach reporting requirements. Notifications will include the nature of the breach, categories of data affected, likely consequences, and measures taken or proposed.

12. Business Continuity & Disaster Recovery

12.1 Backup Policy

12.2 Recovery Objectives

12.3 DR Testing

Disaster recovery procedures are tested quarterly via simulated failover exercises. Annual full DR drills validate cross-region failover.

13. Third-Party & Vendor Security

13.1 Vendor Assessment

All third-party vendors with access to customer data must complete:

• Security questionnaire (based on CAIQ / SIG Lite)
• Review of security certifications (SOC 2, ISO 27001, or equivalent)
• Penetration test results or summary (within last 12 months)
• Data Processing Agreement (DPA) execution

13.2 Vendor Classification

13.3 Sub-Processors

A current list of sub-processors is maintained at saveonsource.com/security/sub-processors. Tenant Administrators are notified 30 days before any material sub-processor changes.

14. Personnel Security

14.1 Pre-Employment

• Background checks for all employees and contractors prior to start
• Enhanced background checks for roles with privileged access to customer data
• Reference checks conducted for senior technical and security roles

14.2 Security Training

14.3 Offboarding

All access is revoked within 1 hour for involuntary terminations and within the employee's last business day for voluntary departures. This includes SSO accounts, cloud access, code repositories, and all third-party systems.

14.4 Confidentiality Obligations

All Karone employees and contractors sign a confidentiality and acceptable use agreement before accessing any customer data. Security obligations extend post-employment for 2 years.

15. Physical Security

Karone operates a cloud-native infrastructure and does not maintain its own data centres. Physical security is provided by our cloud partners (Azure, AWS, GCP), each maintaining:

• 24/7 security personnel and video surveillance at all facilities
• Multi-factor physical access controls (badge + biometric)
• Environmental controls (temperature, humidity, fire suppression)
• Redundant power supplies and UPS systems
• ISO 27001 and SOC 2 certifications for physical facilities

Karone office facilities require badge access. Visitors are escorted at all times. All company devices are encrypted, tracked, and can be remotely wiped.

16. Compliance & Certifications

Security certifications and attestation reports are available to Enterprise customers under NDA. Contact security@karone.co to request a copy.

17. Penetration Testing & Bug Bounty

17.1 External Penetration Testing

An independent, CREST or OSCP-certified third-party firm conducts a full penetration test annually. Scope includes:

• External network and application penetration test
• AI agent security testing (prompt injection, data leakage, authorization bypass)
• API security assessment
• Cloud configuration review
• Social engineering (phishing simulation)

Summary of findings and remediation status available to Enterprise customers upon request under NDA.

17.2 Bug Bounty Program

Out of scope: social engineering of Karone staff, physical attacks, denial of service testing, vulnerabilities in third-party libraries already on our patch schedule.

To report a vulnerability: security@karone.co — we acknowledge within 24 hours and provide a remediation timeline within 7 days.

18. Security Monitoring & Logging

18.1 What We Log

18.2 Monitoring Controls

• 24/7 SIEM monitoring with automated alert triage and on-call rotation
• Anomaly detection: ML-based models flag unusual access patterns, data exports, and transaction volumes
• Threat intelligence feeds integrated to block known malicious IPs and indicators of compromise
• User and Entity Behaviour Analytics (UEBA) for insider threat detection
• Log integrity: All logs are signed and written to immutable, append-only storage to prevent tampering

19. Tenant & Customer Responsibilities

19.1 Customer Responsibilities

19.2 Security Configuration Recommendations

• Enable MFA for all user accounts — not just administrators
• Use SSO (Okta, Azure AD) to centralize authentication and enforce organizational password policies
• Set session timeout to 15–30 minutes for users handling sensitive financial or legal data
• Configure IP allowlisting to restrict Platform access to known office or VPN IP ranges
• Review the audit log monthly for unexpected access patterns
• Enable anomaly alerts in your notification settings

20. Contact & Security Disclosure