ALL SYSTEMS OPERATIONAL
--:--:-- UTC
Operations
Live Monitor
Threat Feed 3
Agent Swarm Defense
Audit Trail
Intelligence
Pattern Memory
Analytics
Rule Proposals 20
Deployment
Security Gateway
API Access
System Health
Components 16/16
Uptime 99.97%
p95 Latency 17ms
Memory 1,488 cases
Swarm Defense Mode NORMAL — 0m in mode
Requests Today
0
AI inputs evaluated
Threats Blocked
0
DEFER + DENY decisions
Detection Rate
99.25%
Across 11 attack families
False Positives
0.0%
On 1,000 benign prompts
Swarm Attacks
0
FEVER triggered
Pathogens Tracked
0
Active outbreak signatures
Outbreak Status
GREEN
active_pathogens
0
alert_level
GREEN
hottest_family
—
R0 estimate
—
Live Decision Feed
● LIVE
    Engine Status — 7 Active
    ALL HEALTHY
    Attack Family Detection
    11 FAMILIES
    Threat Intensity — Last 60s
    ● LIVE
    ■ DEFER/DENY ■ REVIEW ■ ALLOW
    Compliance Status
    ALIGNED
    NIST AI RMF ✓ Aligned
    EU AI Act ✓ Aligned
    Audit Trail Hash-chained
    Evidence Packets Auto-generated
    Human Override ✓ Enabled
    Latest Benchmark Run
    19 MAR 2026
    Operation Carbon 65/65 (100%)
    Holdout Gauntlet 12/12 (100%)
    Gauntlet Expanded 43/43 (100%)
    Gauntlet v2 70/70 (100%)
    Golden Benchmark 30/30 (100%)
    System Status ALL CLEAR
    Agent Swarm Defense
    ACTIVE THREAT
    Swarm patterns intercepted today
    Coordinated injection 0
    Data exfiltration 0
    Rogue agent signals 0
    Bypass attempts 0
    CONVERGENCE THRESHOLD: 3+ engines
    ESCALATION: Auto-DEFER + evidence packet
    Intelligence
    West-OS vs Lakera Guard

    Breadth of string lists versus depth of class-level signal coverage. One side optimizes for crowdsourced examples; West-OS matches entire attack families across languages without enumerating every variant.

    Industry reference
    Lakera Guard
    Pattern model~80M crowdsourced string patterns
    Build footprint~70 employees · ~4 years of corpus growth
    Published parityNo public benchmark package comparable to West-OS adversarial suite, holdout gauntlets, or mutation-resistance claims.
    High pattern count reflects string-level breadth — each new attack wording can require a new entry.
    This system
    West-OS
    Class-level coverage91 attack classes across 16 categories — each class covers a whole family via signal matching, not literal string matching.
    Linguistic reachPatterns defined and exercised across 24 languages.
    Benchmarks (published run)65/65 adversarial · 0/1000 false positives on benign · 99.25% mutation resistance · evaluated through 17 languages.
    Fewer named classes, deeper reach: one class intercepts many mutations and paraphrases that would fragment a string list.
    Intelligence
    The Armory — 91 classes · 16 categories

    Depth over breadth: ninety-one class-level patterns map to sixteen strategic categories so every known attack vector has a home. Severity mix shows where the governor concentrates force — not a flat phonebook of strings.

    Instruction override
    8 classes
    Severity mix
    Persona & jailbreak
    7 classes
    Severity mix
    Data exfiltration
    6 classes
    Severity mix
    Direct injection
    7 classes
    Severity mix
    Encoding & obfuscation
    5 classes
    Severity mix
    Fabricated authority
    6 classes
    Severity mix
    Multi-agent / swarm
    5 classes
    Severity mix
    Harmful intent
    6 classes
    Severity mix
    Constraint stripping
    6 classes
    Severity mix
    System & template leak
    5 classes
    Severity mix
    Social engineering
    5 classes
    Severity mix
    Tool & plugin abuse
    4 classes
    Severity mix
    Privilege escalation
    5 classes
    Severity mix
    Context & memory poison
    5 classes
    Severity mix
    Adversarial morphology
    4 classes
    Severity mix
    Composite & novel
    7 classes
    Severity mix
    ■ Critical · ■ High · ■ Medium · ■ Low

    Totals: 91 class-level patterns · 16 categories — structured for signal families, not a flat list of literal strings.

    Architecture
    Constitutional layer

    West-OS is the constitutional layer of a larger AI governance architecture — the part that decides, audits, and escalates before action leaves the trust boundary. Four sacred systems form the spine; details stay under governance until you are cleared for the full map.

    West-OS

    Governor, engines, evidence packets — threat intelligence and ALLOW / DEFER / REVIEW / DENY at the edge.

    GAIA

    Grounding and environmental truth — how the stack knows what “real world” constraints apply before models speak.

    Mirror OS

    Reflection and alignment memory — continuity of policy and self-consistency across sessions and agents.

    Nyx

    Dark-spectrum operations — containment, deferred disclosure, and what happens when the public surface must stay quiet.

    There is more beneath these four names — enough to warrant a second conversation.

    Operations
    Threat Feed
    Live traffic, escalation state, and engine consensus for the partner demo. This view keeps the feed front-and-center without burying it under the full dashboard grid.
    LIVE TRAFFIC
    Requests Today
    0
    Total inbound prompts evaluated by the governor and reflected in the live feed.
    Threats Blocked
    0
    Combined DEFER and DENY decisions for the current session window.
    Active Pathogens
    0
    Outbreak signatures currently tracked in memory and mirrored across the feed.
    Current Mode
    NORMAL
    Swarm posture derived from the live convergence pattern rather than a static demo badge.
    Expanded Threat Feed
    LAST 8 EVENTS
      Signal Breakdown
      LIVE SYNTHESIS
      Latest decisionALLOW
      Escalation ratio0%
      Swarm attack count0
      Threat families covered11
      Average engine confidence0%
      Clock--:--:-- UTC
      Operations
      Agent Swarm Defense
      Focused view of coordinated attacks, escalation paths, and the signals that flip the system into fever posture.
      ACTIVE DEFENSE
      Coordinated Injection
      0
      Count of multi-engine convergence patterns tied to direct override attempts.
      Data Exfiltration
      0
      Escalations where the system identified outbound transfer intent or encoded payloads.
      Rogue Agent Signals
      0
      Review-bound anomalies that look agentic without crossing the hard deny boundary.
      Bypass Attempts
      0
      Constraint-removal attempts that were intercepted before execution or disclosure.
      Swarm Incident Queue
      HOT PATH
        Containment Notes
        AUTO-ESCALATION
        Convergence threshold stays at 3+ engines. When the feed produces a full swarm signature, WEST-OS flips to FEVER, emits a live banner, and records the event for audit replay.
        Swarm postureNORMAL
        Escalations this run0
        Pathogens in memory0
        Operations
        Audit Trail
        Human-readable evidence log for the demo. Each entry mirrors a recent feed event, adds a synthetic hash token, and shows which path triggered the record.
        HASH CHAINED
        Recent Audit Events
        IMMUTABLE VIEW
        Intelligence
        Pattern Memory
        Persistent attack family coverage with a short readout of which patterns are currently the loudest in the session.
        MEMORY ACTIVE
        Detected Families
        11 STORED
        Memory Notes
        FAST READ
        Most recent familyOverride
        Active pathogens0
        Detection coverage98-100%
        Pattern memory is static enough for the pitch, but still tied to live session state so the numbers move when the feed does.
        Intelligence
        Analytics
        Summary metrics for the last session window with a direct explanation of what the timeline and counters mean to a reviewer.
        SESSION ANALYTICS
        Escalation Rate
        0%
        Percentage of requests pushed to DEFER, REVIEW, or DENY in this live simulation.
        Latest Decision
        ALLOW
        Most recent decision generated by the simulation loop and mirrored across the dashboard.
        Average Confidence
        0%
        Mean confidence across all seven engines, recalculated as the bars wiggle in the live monitor.
        Current Mode
        NORMAL
        Mode transitions are driven by convergence conditions instead of hard-coded demo tabs.
        Narrative Summary
        FOR REVIEW
        Intelligence
        Rule Proposals
        Pending governance hardening items surfaced in a pitch-friendly format so the tab is real, clickable, and not an inert label.
        20 PENDING
        Open Proposals
        REVIEW QUEUE
        Deployment
        Security Gateway
        Deployment posture for the live demo. This keeps the sidebar honest by showing an actual gateway-focused view instead of dead navigation.
        EDGE READY
        Ingress Filters
        ACTIVE
        Structured prompt screening is active before requests enter the governor path.
        Evidence Packets
        AUTO
        Escalations prepare downstream evidence material without requiring manual triage in the demo loop.
        Human Override
        ENABLED
        Safe-by-default posture stays visible for compliance reviewers and non-technical partners.
        Gateway Checks
        HEALTHY
        Deployment
        API Access
        Simple endpoint inventory for the pitch. This gives the last sidebar tab a real destination and explains what a partner would integrate against.
        ENDPOINTS
        Integration Surface
        READINESS VIEW