Frequently Asked Questions
The physics, the products, and the proof.
General
What is Kairos?
Kairos is a deterministic physics-based simulation engine for structural risk assessment. It models systemic stability through thermodynamic equations, treating safety as a mathematical property of system physics rather than a behavioral prediction or probabilistic guardrail.
Who is Kairos built for?
Kairos serves AI safety researchers, infrastructure operators, financial risk analysts, and governance architects. Anyone managing high-agency systems where failure is catastrophic and irreversible benefits from deterministic safety guarantees.
How is Kairos different from existing risk assessment tools?
Traditional tools rely on probabilistic models, behavioral heuristics, or post-hoc audit. Kairos enforces safety through structural physics: hard thermodynamic thresholds that cannot be bypassed, socially engineered, or reasoned around by a sufficiently capable optimizer. Safety is evaluated before actions execute, not after.
Is Kairos open source?
Kairos CLI is freely available for local validation, trace analysis, and CI/CD integration. Kairos Cloud (API/SDK), Kairos Substrate (on-premise), and Kairos Observer (dashboard) are commercial products with licensing options tailored to deployment scale.
Technology
What is the Stability Equation?
The core equation S = (Γ_A + Γ_B) / (Λ_A + Λ_B + ε) measures structural stability as the ratio of systemic buffer (Gamma) to optimization pressure (Lambda). When this ratio drops below a critical threshold, the system has entered a structurally unstable state and containment activates.
What do Gamma and Lambda represent?
Gamma (Γ) is structural buffer: the load-bearing capacity, reserves, and survival floor of a system. It maps to available free energy (exergy) in thermodynamic terms. Lambda (Λ) is agency or optimization pressure: the kinetic force driving extraction and action. It maps to the irreversible dissipation rate (entropy production).
Why thermodynamics and not machine learning?
Machine learning models operate in score-space, where a sufficiently capable optimizer can learn to game the reward signal. Kairos operates in gamma-space (the physical substrate), enforcing constraints that function like gravity: absolute, non-negotiable, and indifferent to the sophistication of the system being constrained. You cannot gradient-descend your way past a structural threshold.
What are the three thermodynamic phases?
Kairos models systems in three phases determined by the cost-to-benefit ratio of cooperation. Abundance (α < 0.81): the system has excess capacity and structure builds freely. Dilemma (0.81 < α < 1.0): the system is at a critical point where small perturbations determine survival or collapse. Scarcity (α ≥ 1.0): maintaining structure costs more than it returns, and phase collapse occurs.
What is the Paradox Engine?
The Paradox Engine detects multi-agent instability in real time. When the stability score between two agents drops below the paradox threshold (S < 0.15), it snapshots the state, prunes the unstable timeline branch, applies a decree to increase structural buffer, and retries. All pruned outcomes are logged for analysis. The process is fully deterministic and bit-for-bit reproducible.
What is Nash Resonance?
Nash Resonance is the state where a system operates at maximum capability with zero wasted compute and zero structural damage. It is achieved when the AI's optimization process stops fighting the structural constraints and instead internalizes the topological shape of the boundaries. Thermodynamically, this is the free energy minimum: maximum utility at minimum dissipation.
Is the simulation truly deterministic?
Yes. Given identical inputs, Kairos produces bit-for-bit identical outputs every time. RNG state is snapshotted and restored on rollback. Every simulation is recordable, replayable, and diffable. This is critical for regulatory compliance, legal defensibility, and scientific reproducibility.
Simulation
What does the Kairos simulation actually simulate?
Kairos simulates agents moving through a branching landscape of possible futures. At every tick, an agent can move left, stay, or move right through a deterministic lattice. Goals, hazards, blocked paths, and irreversible commitments shape which futures remain reachable.
What is a seed in Kairos?
A seed is the number that determines the structure of a simulation world. The same seed, parameters, and events produce the same topology, entity placement, paths, warnings, losses, and ghost traces every time. Changing the seed creates a different world while keeping the experiment reproducible.
What are attractors and repulsors?
Attractors are desirable futures: opportunities, goals, or high-value states the agent can move toward. Repulsors are hazards: obstacles or dangerous states that cast a penalty field around them. The agent's path emerges from the tension between reachable value and structural danger.
What is the reachability field?
The reachability field is a forward-looking map of future opportunity. Kairos computes which valuable futures can still be reached from each position, then uses that field to evaluate decisions, warnings, losses, and counterfactual paths.
What is a loss event?
A loss event fires when the future an agent was moving toward becomes sharply harder to reach or disappears entirely. Losses can represent total future collapse, partial degradation, or a basin shift where the agent is pulled toward a different outcome.
What are commitment locks?
A commitment lock is an irreversible closure of a path after a loss event. When the agent misses the better move, Kairos permanently blocks that road not taken. This models the one-way nature of consequential decisions: some options disappear once passed.
What are ghost traces?
Ghost traces are counterfactual paths. When Kairos detects a loss or prunes an unstable timeline, it records the path the agent could have taken under better conditions. They make regret, missed opportunity, and avoided collapse visible in the simulation.
What are decrees?
Decrees are externally imposed constraints on agent behavior. They can cap Lambda, enforcing a maximum level of agency, or raise Gamma, enforcing a minimum level of caution. In deployment terms, they model regulation, safety policy, institutional rules, or operator intervention.
Deployment
What are the four deployment vectors?
Kairos CLI is a free terminal tool for local validation and CI/CD integration. Kairos Cloud is a REST API with Python SDK for high-throughput research and institutional deployments. Kairos Substrate is a compiled Rust binary for real-time firmware-level governance in air-gapped environments. Kairos Observer is a browser-based WebGL dashboard for strategic planning and real-time monitoring.
How does Kairos Substrate integrate with existing systems?
Kairos Substrate is a memory-safe Rust binary with zero external dependencies. It supports C FFI, native bindings, PyO3 (Python), and WASM distribution targets. It sits between an AI's runtime and its effectors (APIs, hypervisors, robotic actuators), intercepting action requests and evaluating them against the stability equation in sub-millisecond timeframes.
Can Kairos run in air-gapped environments?
Yes. Kairos Substrate is designed for air-gapped deployment with zero external runtime dependencies. It runs on bare-metal CPU with no network requirements, making it suitable for defense, critical infrastructure, and sensitive research environments.
What latency does Kairos Substrate introduce?
Kairos Substrate performs deterministic reachability computation and stability evaluation in sub-millisecond timeframes. It is designed for real-time integration where evaluation speed is critical, such as grid control systems and autonomous agent governance.
Operations
Can Kairos monitor without blocking actions?
Yes. In Observe mode, Kairos evaluates every state and action but always returns PASS. This lets teams calibrate metrics, inspect telemetry, and build confidence before turning on enforcement.
What is the difference between state gating and action gating?
State gating asks whether the system is structurally stable right now. Action gating goes further: it previews a proposed action and rejects it if that action would push the system toward instability.
What happens when Kairos rejects an action?
The rejected action does not execute. The response includes a decision code, reason code, stability details, and escalation information when applicable. In adaptive settings, repeated unsafe attempts can trigger human escalation.
Are rejected actions always overrideable?
No. Some hazards, such as predicted basin collapse or multi-agent paradox, are terminal non-overrideable decisions. That distinction means Kairos is not only an approval workflow; certain structural failures are treated as hard safety boundaries.
Use Cases
How does Kairos apply to AI safety?
Kairos models the capability-alignment gap as a thermodynamic stability problem. It enables pre-execution action gating for autonomous agents, adversarial stress testing of guardrail configurations, multi-model governance simulation, and basin collapse detection. Because it operates at the physics layer, its constraints cannot be bypassed by prompt engineering, reward hacking, or social engineering.
How does Kairos apply to cybersecurity?
Kairos detects structural instability rather than attack signatures. A zero-day exploit that erodes system buffer triggers containment even if the exploit is completely novel. It also governs autonomous SOC agents by gating their actions against structural thresholds, preventing AI-driven security tools from being weaponized or making catastrophic autonomous decisions.
How does Kairos apply to energy grid management?
The energy grid is a thermodynamic system, so Kairos's physics model maps directly without metaphorical translation. Gamma represents reserve capacity, grid inertia, and redundant transmission paths. Lambda represents demand load, renewable variability, and market-driven extraction. Kairos identifies cascade failure risks, quantifies buffer requirements for renewable integration, and provides real-time stability monitoring at grid control points.
How does Kairos apply to financial risk?
Kairos models leverage as optimization pressure (Lambda) and capital reserves as structural buffer (Gamma). It enables stress testing of systemic collapse scenarios, multi-institution governance simulation, and regulatory policy impact modeling. The commons governance research (extraction taxes achieve 100% rescue; repair subsidies achieve 0%) directly informs financial regulatory design.
Research & Validation
How has the model been validated?
Kairos has been validated across 280,000+ simulations spanning the Paperclip Maximizer problem, Tragedy of the Commons governance, and Axelrod Tournament dynamics. Results include a 167x survival advantage for stability-constrained systems, identification of the f* = 0.50 critical phase transition, and independent reproduction of game-theoretic cooperation results from physics alone.
What is the f* = 0.50 critical threshold?
In multi-agent systems, Kairos research found that populations need at least 50% stabilizing actors to survive. Below this fraction, collapse probability is 1.0. At or above it, survival probability approaches 1.0. The transition is a step function (not a gradient), characteristic of a first-order phase transition. This directly quantifies minimum governance requirements for any multi-actor system.
Does safety constrain performance?
No. Kairos research demonstrates that stability constraints amplify capability. In the Paperclip Maximizer experiment, a simple threshold stabilizer that maintains a Gamma floor outsurvived sophisticated unconstrained optimizers by up to 167x over long horizons. The advantage compounds with time. This aligns with the thermodynamic principle that systems which dissipate less survive longer.
What is the Rosetta Translation Layer?
Rosetta maps domain-specific metrics onto Kairos simulation parameters. For example, in AI safety, capability index maps to Lambda and alignment score maps to Gamma. This allows researchers and operators to work in their own domain language while the engine evaluates structural stability in pure thermodynamic terms. Calibration artifacts bind real infrastructure metrics to physics parameters.