KALKI-1.5

KALKI AI — Security, Governance & Responsible AI Blueprint

1. Defensive Cybersecurity & Operational Scope

KALKI AI operates within a strict ethical and defensive cybersecurity framework.

Permitted Defensive Capabilities

• Vulnerability Auditing: Automated SAST/DAST assessment on explicitly authorized owned codebases and infrastructure.
• Security Log Analysis: Real-time SIEM anomaly detection parsing syslogs, NetFlow, and CloudTrail event streams.
• Compliance Reporting: Continuous benchmarking against NIST SP 800-53, ISO 27001, and SOC2 Type II controls.
• Incident Response Assistance: Automated playbooks for isolating compromised nodes and revoking token credentials.

Strictly Prohibited Actions

• Creation or distribution of malicious payload binaries or zero-day exploits.
• Unauthorized scanning, brute-forcing, or credential harvesting.
• Bypassing user consent or surveillance of non-consenting parties.

---

2. Role-Based Access Control (RBAC) Matrix

Role	System Config	Knowledge Upload	Agent Execution	Audit Log Access	Admin MFA Required
SuperAdmin	Full	Full	Full	Read/Export	Yes
SecurityAuditor	Read-only	None	Read-only Trace	Full Read	Yes
EnterpriseUser	None	Own Specs	Standard Tasks	Own Logs	Optional
EdgeNode	Read-only Sync	None	Local SLM Only	Telemetry Push	N/A (Mutual TLS)

---

3. Human-in-the-Loop (HITL) Workflow

High-impact actions require explicit human approval before execution.

graph TD
    AgentAction["Agent Proposed Action"] --> GuardCheck{"Action Risk Assessment"}
    GuardCheck -->|Low Risk (Read/Search)| AutoExec["Execute Automatically"]
    GuardCheck -->|High Risk (DB Drop, API Pay, Code Commit)| HITLQueue["Human Approval Queue"]
    HITLQueue --> UserPrompt["Send Interactive Approval Modal"]
    UserPrompt -->|Approved| AutoExec
    UserPrompt -->|Rejected| AbortAction["Abort Task & Log Rejection"]

This site is open source. Improve this page.