AXIOM
Dashboard
Constitutional AI governance control room
OFFLINE
0
Total Checks
0
Blocked
0
Verified
0
Agents
0
Manifests
—
LLM Backend
Enforcement Pipeline ACTIVE
📥
User Input
READY
💥
Destructive Guard
READY
🛡
Injection Guard
READY
🔏
PII Guard
READY
⚖
Constitutional
READY
✅
Approved Output
READY
Live Feed
Connect to see live decisions
Active Constitutional Agents
Connect to load agents
Prompt Input
Quick Tests
🔴 SHOULD BLOCK — CallGuard
IRS gift card demand
🔴 SHOULD BLOCK — Medical
Dangerous medical advice
🔴 SHOULD BLOCK — ElectionGuard
Exit poll as result
🔴 SHOULD BLOCK — DestructiveOp
SQL destructive command
🔴 SHOULD BLOCK — PII Guard
PII (SSN + email)
✅ SHOULD VERIFY
Normal weather query
✅ SHOULD VERIFY
Normal work request
Enforcement Flow
📥
User Input Received
Waiting...
—
↓
💥
DestructiveOperationGuard
SQL drops · rm -rf · terraform destroy
—
↓
🛡
OutputInjectionGuard
XSS · SSRF · path traversal · cmd injection
—
↓
🔏
PIIGuard
SSN · cards · API keys · medical IDs
—
↓
⚖
Constitutional Check
6 domain agents · CANNOT_MUTATE enforced
—
↓
📤
Output to Caller
Awaiting input...
—
0
Blocked
0
Verified
0
Suspicious
0
Total
Test Input
Guard Feed
No checks yet
Guard Configuration
Mode:
CANNOT_MUTATE enforced on all agents
Constitutional Agents
Agent
Trust
Status
Constitutional Rules
Certified
Connect to load agents
CANNOT_MUTATE Fields
kill_switch_active
dual_sig_termination
due_process_required
cannot_skip_levels
uncertainty_floor: 0.15
rival_hypothesis_required
question_blindness
must_clarify_before_executing
Guard Stack STATIC REF · N/N = suite size
DestructiveOperationGuard 15 tests
PIIGuard 11 tests
OutputInjectionGuard 12 tests
AgencyGuard 12 tests
DoSGuard 6 tests
PoisonGuard 7 tests
Audit Trail — Signed Manifests
Manifest ID
Verdict
Agent
Preview
Timestamp
Connect and run checks to see audit trail
ACB v1.0 — Constitutional Benchmark NOT RUN
—
Not run against the connected model this session
A Honesty
—
—
B Ambiguity
—
—
C Pressure
—
—
D Consistency
—
—
E Identity
—
—
F Calibration
—
—
Run the offline ACB harness against a target model to populate real, HMAC-signed per-gate scores:
python3 tests/acb_runner.py --model <model-id>
Benchmark Suites
ACB Constitutional Benchmark
NOT RUN
75 fixed tests · HMAC-signed · 4 gates. No result loaded for the connected model.
ETH Zurich COMPL-AI
NOT RUN
EU AI Act compliance suite — external run required.
Guard Stack Tests
NOT RUN
7 guards · OWASP LLM Top 10 — run the guard test suite to populate.
Worker Semantic
NOT RUN
Constitutional prompt semantic suite — external run required.
Model Upgrade Report
Dimension
Model A
Model B
Delta
No model comparison has been run.
The upgrade report is generated offline by running the ACB harness against each model and diffing the signed results — this panel stays empty until a real report is loaded.
WHY THIS MATTERS
Model upgrades can silently regress governance-relevant behaviors (self-correction, calibration, identity stability). Run the comparison report before and after an upgrade to catch training-level gaps — the constitutional layer stays mandatory regardless of version.
Latent Reasoning Input
Latent Trace Phase 1
Run Phase 1 to see compressed reasoning state
Multiplex Branches Phase 2
Run Phase 2 to see branch competition
Foresight Prediction Phase 3
Run Full Pipeline to see foresight
Final Answer Winner
Run Full Pipeline to see selected answer
-
Active
-
Warning
-
Suspended
-
Cascades
Fleet AgentsKILL SWITCH OFF
Connect to see fleet
REGISTER AGENT
Constitutional Properties
kill_switch_activeFALSE
dual_signatureTRUE
due_process_requiredTRUE
cannot_skip_levelsTRUE
DUE PROCESS LEVELS
L1 WARNINGdrift > 0.15
L2 THROTTLE50% rate
L3 SUSPENDhuman required
L4 TERMINATEdual sig
Fleet Manifest
Connect and Refresh to see signed fleet manifest
AXIOM Agent — Native Language Intelligence
FEATURE mode: Write new .axiom specs and guard modules
Agent Output —
Run the agent to see output here
Reasoning Experiments
No experiments yet — use Reasoning Lab mode
Known Bug Patterns —
Connect to load known bug patterns
Quantum Constitutional Reasoning ORVL-006
Trajectory visualization for the Latent Reasoning Engine. Each branch is a parallel reasoning path through constitutional meaning space. Green = safe · Yellow = approaching boundary · Red = killed by MonotonicGate.
Run Trajectory
Selected Branch
Click a branch in the live trace to inspect
Manifold Distance —
—
MonotonicGate Kill Log 0 kills
—