v5.0 — AI Safety Platform · 8-Phase Defence · 31 MCP Tools · SDK v0.2

    AI Safety & Governance
    for Regulated Industries.

    Enterprise AI safety and governance on your infrastructure. Cryptographic audit chain, 7 compliance frameworks, champion/challenger registry, and agent skill & MCP server governance — combined with a defence-in-depth safety platform: ML classifier ensemble, agentic tool guards, conversation monitoring, and a red-team certification runner. EU AI Act, SR 11-7, NIST AI RMF, and four more — enforced, not reported.

    Planner
    Retriever
    Researcher
    Analyst
    Coder
    Critic
    community-mcp422 · blocked at gate
    Aegis
    LangGraph
    CrewAI
    AutoGen
    Blocked at gate

    Your agent mesh, governed — every agent, skill, and MCP server registered, scanned, and policy-gated. Unsafe artifacts never reach production.

    AI Safety Platform v2

    Safety built into every
    request, handoff, and promotion.

    A mandatory 8-phase defence-in-depth pipeline — ML classifier ensemble, agentic tool guards, cross-turn conversation monitoring, trust-boundary delegation, safety SLOs, red-team certification, and a registry clearance gate — enforced before any response reaches your users.

    18+adversarial probes
    4ML safety classifiers
    < 2msregex pre-screen
    Explore AI SafetySafety API Docs

    Professional and above · Red-team certification and safety analytics — Sovereign only

    8-Phase Defence Pipeline

    1
    Policy EngineRegex + ML
    2
    Agentic GuardTool validation
    3
    PII MaskingPresidio
    4
    ConversationCross-turn
    5
    Trust BoundaryDelegation
    6
    SLO EngineBurn-rate
    7
    Red-Team CertProbe corpus
    8
    Clearance GateRegistry gate
    Safety Policy Engine — Demo
    Illustrative
    HIGH

    ABUSE-001

    adversarial_abuse · tool-agent

    blocked
    HIGH

    JAILBREAK-001

    jailbreak · researcher-agent

    blocked
    MED

    PRIVACY-002

    privacy_leakage · coder-agent

    auto-redacted
    CRIT

    HARMFUL-001

    harmful_content · planner-agent

    blocked
    CRIT

    OVERSIGHT-001

    human_oversight · executor-agent

    review-queued
    Regex rules + ML ensembleavg latency 2 ms
    Safety SLOs — Error Budgets
    PII protection
    0 leaks0.2× burn
    99.7% budget remaining
    Block-rate budget
    ≤ 10%0.7× burn
    98.4% budget remaining
    Injection budget
    ≤ 250.1× burn
    100% budget remaining
    Safety posture score
    > 901.1× burn
    94% budget remaining

    Burn rate > 14.4× triggers PagerDuty page automatically

    ML Classifier Ensemble
    Multi-Agent Trust Boundary
    Cross-Turn PII Memory
    Approvals & Human Review
    Red-Team Probe Corpus
    Multimodal Image Safety
    ModelSafetyRegistry Clearance
    Automated Safety Reports
    Conversation Monitor
    Error-Budget Burn-Rate SLOs
    Red-Team Certification
    Audit-Chain Anchoring
    Operational Pulse

    Governance Telemetry

    What the platform's telemetry looks like in a running workspace — illustrative demo data. See it live with real events in the hosted sandbox.

    Safety Policy Engine

    Active

    llama_guard + openai_moderation

    Red-Team Cert

    42 / 42

    probes detected · certification passed

    Human Review Queue

    3 pending

    SLA: 15 min critical

    Safety SLOs

    4 healthy

    burn rate < 14.4× on all

    Gateway Latency

    Demo
    30s25s20s15s10s5s0s30ms37ms44ms55ms42ms SLA

    Platform Activity Map

    Model Registry
    LLM Gateway
    Audit Chain
    Policy Engine
    Safety Engine

    Event Stream

    DEMO FEED
    13:42:44mcp_server.registered — internal-docs-mcp — scan passed — token auth · signed
    13:42:44llm.budget.alert — team: ml-ops — 82% of monthly cap reached
    13:42:44llm.gateway.blocked — jailbreak attempt — prompt-safety policy
    13:42:44safety.trust.boundary — researcher→coder — permission escalation blocked
    13:42:44audit.chain.verified — 247 events — chain intact

    Deploy Without Borders

    One Universal Helm chart. Every cloud. Your infrastructure.

    helm — Azure (AKS)
    $ # From the Aegis Sovereign release bundle — AKS
cd aegis-sovereign/helm

helm upgrade --install aegissovereign \
  ./sovereign-gateway \
  --namespace aegissovereign --create-namespace \
  --values sovereign-gateway/values-production.yaml

    The sovereign-gateway chart ships in every release bundle. A public Helm repository lands with the marketplace listings.

    PII Masking Events:2,847,424
    Compliance Evals:38,205
    Safety Violations Blocked:14,874
    Demo — illustrative data