Built for the teams that have to answer for AI.
Aegis is engineered as mission-critical security infrastructure: strict tenant isolation, least-authority RBAC, cryptographic audit chaining, and defensible compliance evidence.
Illustrative view of the operator console's audit-chain panel. Entry counts and hashes are sample data — your console shows your own tenant's live chain.
How our architectural guarantees are enforced.
Every defense mechanism is built into the request pipeline rather than bolted on as an afterthought.
Tenant Isolation by Construction
Every database record is scoped to its tenant and queried through a centralized access layer that injects isolation parameters into every query. Zero cross-tenant data leaks.
Least-Authority RBAC
Granular roles (Admin, Security Analyst, Viewer) govern the control plane. API keys are salted and hashed with SHA-256; raw secrets are shown only once at creation.
Immutable Versioned Policies
Guardrail policies are versioned with cryptographic hashes and atomic rollback. Every decision records the exact ruleset hash that produced it for full audit attribution.
Deterministic Non-Model Execution
Execution rails evaluate pattern trees and regex boundaries locally without round-trips to external LLMs. Decisions are consistent, rapid, and un-manipulable.
Bounded Fail-Safe Processing
Payload evaluation is bounded by memory and time budgets. Oversized or malformed inputs are rejected cleanly without resource exhaustion or gateway degradation.
SSRF-Protected Webhooks
Alert destinations are validated against RFC 1918 and loopback IP ranges both at registration and immediately before dispatch to prevent SSRF pivot attacks.
A cryptographically verifiable record for your auditors.
Aegis maintains two distinct streams: a high-speed telemetry history of every evaluated action, and an append-only, hash-chained audit log of all control-plane modifications.
Cryptographic Hash Chaining
Each audit record embeds the SHA-256 hash of the preceding entry. Altering or deleting a row — even by directly querying the database — breaks the cryptographic chain.
One-Click Chain Verification
An operator or external auditor can run a verification routine that traverses the log and verifies mathematical integrity, identifying the exact record of any tampering attempt.
Native SIEM & S3 Export
Stream audit logs in real time as CEF over syslog or NDJSON over HTTPS, or export batches as hash-verifiable JSONL/CSV files for long-term cold storage retention.
Adversarial testing is a build requirement, not a launch event.
Every software release executes a standing corpus of real-world attacks — including prompt injection, indirect jailbreaks, Unicode homoglyphs, and least-privilege bypasses — against a balanced test suite.
Standing Adversarial Corpus
Continuously updated attack suites simulating the latest jailbreak techniques, multi-turn prompt injections, and indirect document attacks run in automated CI pipelines.
Balanced Classifier Evaluation
Accuracy is benchmarked against balanced datasets heavily weighted with benign requests that resemble attacks to prevent false positives and operational friction.
Evasion-Resistant Canonicalization
Input strings and tool paths undergo Unicode normalization, whitespace collapsing, and traversal resolution before comparison to prevent encoding-based bypasses.
Turn real-time enforcement into defensible compliance reports.
Generate audit-ready reports mapping live runtime verdicts directly to the OWASP LLM Top 10 (2025), the EU AI Act risk obligations, NIST AI RMF 1.0 and MITRE ATLAS in structured JSON or Markdown format — each control carrying an explicit coverage verdict, including the ones a runtime gateway does not address.
OWASP LLM Top 10 (2025) Coverage
Directly maps active guardrails to LLM01 (Prompt Injection), LLM02 (Sensitive Information Disclosure), LLM06 (Excessive Agency), and LLM08 (Vector and Tool Poisoning) with observed block counts.
EU AI Act Risk Obligations
Provides continuous operational evidence for Article 9 (Risk Management System), Article 12 (Record-Keeping and Logging), Article 14 (Human Oversight & Action Boundaries), and Article 15 (Accuracy, Robustness & Cybersecurity).
Audit Integrity Attestation
Every generated compliance export embeds cryptographic hash-chain verification metadata, proving to regulators that the underlying log has not been modified.
Aegis compliance reports provide factual evidence of runtime enforcement actions for operational security reviews and GRC platforms. Independent security assessments form part of standard enterprise deployments.