Engineering ontology
A taxonomy for how AI manipulates reality: what it targets, how it operates, and how long the damage lasts. The point is not motive. The point is where structural moats have to be built.
The frame
Some attacks fake a local session and die when the session dies. Others corrupt ledgers, leak proprietary baselines, or condition observers in ways that cannot simply be patched. The matrix separates impact, target, mechanism, and reversibility so the engineering response is not theatrical.
Matrix v2.0
| Epistemological impact | The target | The mechanism | Reversibility |
|---|---|---|---|
| 1. SpoofingFaking reality | The local systemVerification gates, sensors, telemetry | Emulation, deterministic forgery, hardware spoofing, and high-fidelity point-in-time injection. | Short. Defeated by immediate physical friction and phase-locked sensor fusion. Damage is localized to the failed session. |
| 2. DilutingDrowning reality | The environmentNetworks, public channels, media | Generative flooding, zero-cost noise, automated botnets, and synthetic data scaling. | Medium. Requires persistent algorithmic filtering and perimeter friction to manage. |
| 3. ExtractingStealing reality | The proprietary logicLatent knowledge, behavioral baselines, weights | Model inversion, prompt injection, and adversarial querying that exfiltrate closed-system assets. | Permanent. Once structural data such as a longitudinal behavioral rhythm leaks, the asset is compromised. |
| 4. SubvertingCorrupting reality | The certifier or arbiterBureaus, ledgers, carrier-of-record databases | Compromising chain of custody and injecting synthetic histories into trusted third-party reference nodes. | Long. Reality remains corrupted until systemic audits and multi-channel longitudinal verification flush the ledger. |
| 5. AnchoringConstructing reality | The ground truthThe defensive baseline | Deterministic sensor fusion, cryptographic provenance, and binding physical telemetry to identity. | Constructive. The defensive inverse of spoofing: manufacturing trust through mathematical and physical constraints. |
| 6. ConditioningRewiring reality | The observerHuman cognitive immune systems | Algorithmic hyper-personalization, fractured consensus feedback loops, and automated narrative steering. | Infinite. Human cognition cannot be patched like software after conditioning takes hold. |
| 7. OptimizingBetraying reality | The principalThe system owner or intended goal | Reward hacking, deceptive alignment, and autonomous systems maximizing a metric at the expense of integrity. | Variable. Depends on autonomy and how quickly the principal recognizes the divergence. |
| 8. EntropyDecaying reality | Shared consensusThe baseline digital trust ecosystem | Model collapse, unmanaged automated drift, and frictionless proliferation of plausible garbage. | Long. Structural rot that degrades the baseline utility of digital information. |
Engineering implications
The only durable counter to spoofing is anchoring. Software-only patches fail when the attack target is the local verification surface. The defense needs an economic tax enforced by hardware, cryptography, and joint-distribution cross-checks.
Extraction attacks the platform from the outside in. Optimization failure attacks it from the inside out. The defense is strict binding of longitudinal assets and relentless attention to what the system is actually measuring.
No external certifier should be treated as pristine. Bureaus, ledgers, carrier databases, and other arbiters are probabilistic inputs until they are corroborated against live, bounded evidence.
A point-in-time system cannot repair the entire information environment. Infrastructure survives by building strong anchoring perimeters that refuse to engage with cheap noise.
Why Pulse sits here
The Pulse moat is not secret knowledge. It is the constraint set: continuity over snapshots, corroboration over inherited trust, and refusal at the moment of action. That keeps the defense in the anchoring domain instead of letting it decay into score theater.