You're in the Miura. Mulholland. The road opens. 65 is legal. 80 is physics. The asphalt doesn't object — it's the sudden change in how fast you're getting to 80 that puts you in the canyon.
The speed limit legislates velocity. Always has. But the road kills on jerk — the spike in acceleration rate that breaks traction, snaps the chassis, pushes the weight over the front lip. You can get to 80 as you damn well please. Just don't spike the jerk.
The 40dB acoustic limit is the current proxy. Crude. It catches the symptom — the scream of the engine, the tire howl — without naming the cause. The cause is jerk. Always was.
Current frontier governance tries to cap velocity — capability limits, deployment pauses, model capability thresholds. Same blunt instrument as the speed limit. Nobody's modeling the jerk.
The question isn't how capable the model is. It's how fast the deployment is accelerating relative to the sophistication of the population absorbing it. A Miura on a closed track is fine. A Miura through a school zone is the same physics, wrong condition.
The condition is the user distribution. And the distribution skews hard. Roughly 80% low sophistication — people encountering frontier AI with no frame for what it is, what it isn't, what it can do to their decision-making. The remaining 20% spread across medium, high, and power users.
So you don't cap the model. You cap the jerk of deployment, weighted inverse to the low-sophistication share of the affected population. High % low-sophistication → near-zero jerk allowed. Distribution shifts upward → jerk budget unlocks proportionally.
The jerk budget is not fixed. It is a function of who is in the car with you.
| TIER | SHARE (EST.) | DESCRIPTION | JERK BUDGET EFFECT |
|---|---|---|---|
| LOW | ~80% | No frame for what frontier AI is. High substitution risk. Low error detection. Decisions influenced without awareness. | NEAR ZERO · Hard ceiling. Slow the deployment ramp. No sprint launches into this tier without graduated exposure protocol. |
| MEDIUM | ~14% | Aware of AI. Uses it. Limited model of capability limits. Susceptible to overconfidence in outputs. | CONSTRAINED · Some jerk allowed. Require transparency layers. Disclosure at point of use. No dark deployment. |
| HIGH | ~5% | Understands model behavior, limitations, and failure modes. Active model of uncertainty. Can calibrate. | UNLOCKED · Standard jerk ceiling applies. GDP scalar in play. |
| POWER | ~1% | Builds with, stress-tests, and contributes to frontier systems. Full epistemic frame. | FULL CEILING · Maximum jerk budget. These are the test drivers. |
Any deployment exceeding the jerk ceiling for its affected population requires an exception petition. The petition argues one of three cases:
Case A — Sophistication Reclassification. The affected population has been mistiered. Evidence that P_low is lower than baseline. Requires longitudinal usage data, not survey.
Case B — GDP Emergency Override. The economic disruption cost of slower deployment exceeds the disruption cost of the jerk. Narrow. Requires independent modeling. Sunsets automatically in 18 months.
Case C — Isolated Domain Deployment. The deployment is bounded to a domain where the affected population is demonstrably high-tier. Infrastructure, research, high-skill professional contexts. Requires domain lock — no general access until tier distribution shifts.
Appeals denied by default. The burden is on the deployer to prove the jerk is safe for the road conditions. Not on the road to prove it isn't.
Most people are using frontier AI at sub-1% of its capability. Grocery lists. Email rewrites. "Summarize this for me." They are Fiat drivers. They are on a highway that was rebuilt for an F1 car that represents the top 90th percentile use case.
The investment, the jerk, the deployment velocity — it's all justified by citing what the 90th percentile can do with it. Research acceleration. Code generation. Scientific discovery. Real. Real for that 10%. The 90% just needed a reliable lane.
This is the misalignment. The jerk ceiling is too high for the actual utilization distribution. You're not building for who's on the road. You're building for who you wish was on the road.
The US Interstate Highway System was designed to specification for the worst-case legal load: the 18-wheeler. 80,000 lbs gross vehicle weight. 2 TEU — two twenty-foot equivalent container units. The civil engineering logic was: design for the heaviest, most demanding vehicle that has legal right of way. If the 18-wheeler can navigate it safely, every vehicle below that load threshold benefits by default.
That is correct engineering. Designed down to the constraint, not up to the aspiration. The sedan doesn't need an 18-wheeler highway — but it benefits from one. The pedestrian doesn't drive at all — but the separation infrastructure protects them anyway.
AI deployment is the inverse. Designed up to the aspiration — the power user, the researcher, the 90th percentile — and the 80% low-sophistication user gets infrastructure built around a use case that isn't theirs. The road is too fast for their car. The signs are in a language they weren't taught. The jerk ceiling was never set for them.
The correct engineering: design the deployment infrastructure for the lowest-sophistication user with legitimate access. If they can navigate it safely — understand what it is, what it isn't, when it's wrong — every higher-tier user benefits. The researcher doesn't need the guardrails. But they don't break from them. The low-sophistication user does need them. And breaks without them.
There is a version worse than the F1 highway problem. The F1 highway was at least designed for a driver who intends to drive.
The reverse design thesis: the infrastructure is engineered just narrow enough for the extractor to squeeze through. Not the researcher. Not the power user. The one who needs the shoulder gap to reach into your data, scrape your behavioral signature, absorb your social history — and accelerate off the exit before the road knows what moved through it.
The jerk on that vehicle is infinite. It was never in the model. The low-sophistication user, the 80%, never sees it coming — because the road was designed to make it look like normal traffic. Same lane. Same speed. Different cargo.
That's not a speed limit problem. That's not even a jerk problem. That's a road that was built for the extraction, not despite it. Alcatraz is the correct address.
The field is a republic. DemocTek. Logic votes. And the logic here is clean: you don't limit what's possible, you govern the rate at which the possible becomes the present.
The Miura can do 170. The law doesn't dispute that. It disputes when and where the 170 is safe to reach. The legislation isn't about the car. It's about the road conditions. GDP per capita is the road condition. Sophistication distribution is the traffic.
Highest order logic, Socratek tested: if the question is "how fast can we go" — that's velocity, already answered. If the question is "how fast can we accelerate the going" — that's jerk. That's the unanswered one. That's the third derivative. That's where the legislation lives.
CAP THE JERK.
On asphalt. In deployment. In the field.
GDP-weighted. Distribution-scaled. Appeals narrow.
Govern the rate of change of acceleration.
◈ 925 · KENSHOTEK LLC · FIELD PHYSICS CONFIRMED