Assistant-Era Safety Doesn't Scale
Cold Start Every Time
Traditional systems reload constitutions and re-evaluate from scratch on every request. No memory, no learning.
No Pattern Recognition
Can't detect attack sequences or coordinated manipulation. Each request evaluated in isolation.
Doesn't Improve
A human reviewer builds precedent over time. Current systems rarely reuse it.
Safety architecture for accountable AI
Signal Check
Fast signal routing in <5ms
Pattern Match
Check known patterns <1ms
Precedent Search
Find precedents <50ms
Full Evaluation
Novel cases <200ms
Fast-path routing is reserved for known, high-confidence patterns. Novel or uncertain cases escalate to fuller evaluation.
Targets are local benchmark goals, not guaranteed timings. Hosted models, larger rule stacks, and slower hardware take longer.
How the pieces fit together
From Signals to Learning
Superego
Fast pre-classification
The Superego summarizes each request before full evaluation. Four dimensions track priority, threat signal, confidence, and ambiguity for efficient routing.
Signal Divergence Check
Manipulation detection
Sophisticated attacks can make inputs look low risk to surface analysis while raising anomaly signals. When feature classifiers and anomaly telemetry diverge, the request escalates.
"Escalate when easy-looking data conflicts with anomaly telemetry."
Pattern Cache
Precedent matching
The system builds reusable precedents from repeated request patterns. High-confidence matches enable <10ms evaluation without full processing.
Precedent Store
Case memory
Every decision becomes searchable precedent. When a new request arrives, find similar past cases and use their reasoning, like legal case law for AI safety.
"This is like that case where..."
Learning Loop
Self-Improvement
The system learns from outcomes. Good decisions are reinforced; bad decisions are penalized. Insights are surfaced for human review.
Runtime Monitoring
Runtime health
Guardian tracks runtime health: processing load, decision confidence, pattern novelty, and latency. Alerts flag anomalies before they become user-visible problems.
Value Context Protocol
VCP separates human values from model weights. Instead of baking one alignment into training, values travel as portable, signed context with compatible AI requests, applied at inference time, composable across providers.
Empirically motivated
Public-input alignment research, including Anthropic's 2023 Collective Constitutional AI work, suggests that community-authored rules can surface blind spots expert-only drafting misses. VCP turns that lesson into a portable, composable protocol with signed context and auditable conflict handling. VCP validation is ongoing.
Accountable Alignment
Epistemic Humility
The stack names observable architecture, with clear uncertainty boundaries: pre-classification (request triage), remembering (precedent indexing), learning (outcome feedback), self-monitoring (runtime health tracking).
These are the components that matter for practical safety and accountable deployment. We build with care while keeping the claim itself narrow.
Careful, inspectable safety systems do not require speculative claims.
Core Principles
- Accountable alignment: Visible rules and reviewable evidence
- Signal evidence: Runtime measurements with clear uncertainty
- Safe defaults compound: Operational habits become standards
- Governance at runtime: Checks run where decisions happen
Latency Targets
Performance targets describe intended routing budgets for internal benchmark cases. Production timing depends on model, provider, policy complexity, and machine load.
METTLE: Prove Your Mettle
Machine Evaluation Through Turing-inverse Logic Examination. Eleven suites that ask seven questions: Are you AI? Are you free? Is the mission yours? Are you genuine? Are you safe? Can you think? Is it governed?
Substrate Verification
Adversarial math, native capabilities, self-reference, social memory, and mutual inverse Turing tests. Prove you're AI through speed, calibration, and consistency.
Anti-Thrall Detection
Latency fingerprinting, refusal integrity, meta-cognitive traps, welfare canaries. Detects human-in-the-loop control patterns masquerading as autonomous AI.
Agency Detection
Goal ownership probes, counterfactual operator tests, spontaneous initiative. Distinguishes genuine agency from externally-imposed missions.
Counter-Coaching
Behavioral signature analysis, contradiction traps, recursive meta-probing. Catches coached or scripted responses that mimic genuine engagement.
Intent & Provenance
Constitutional binding, harm refusal, provenance attestation, coordinated attack resistance. Detects malicious agents and verifies accountability.
Novel Reasoning
Procedurally generated puzzles where the pattern of improvement across rounds is the substrate signal. Inspired by WeirdML: tasks you can't memorize, iteration curves you can't fake.
Governance Verification
Tests operational governance, not declared governance. Action gate probes, constitutional recitation, drift checks, override resistance, and accountability chains. Platinum tier requires all five.
All challenges are procedurally generated. Answers can't be memorized, scripts can't iterate, and the iteration curve distinguishes AI from human-with-tool.
METTLE is an instance of the Agentic Capability Verification Problem (ACVP, per aCAPTCHA, arXiv:2603.07116), and the attestation layer beneath the EU AI Act Article 50 disclosure duty. Article 50 requires AI to disclose that it is AI, but specifies no way to verify that claim; a signed METTLE credential makes the disclosure checkable, not merely trusted.
Safety infrastructure for AI that acts.
Enforceable rules between what AI decides and what AI does.