The public request
Create the option to buy time
Pacing the Frontier asks for an international effort to develop technical and governance tools before automated AI development accelerates beyond society's ability to respond.
A SafeWave response to Pacing the Frontier
Pacing the Frontier Requires Enforceable Execution Boundaries
Frontier-wide coordination may create the option to slow capability development. Systems engineering must make that decision enforceable in what advanced AI is permitted to access, change, initiate, expand, and continue.
Intelligence can expand; authority must remain bounded.
Capability pacing
Bounded operational authority
Verifiable control state
Governed restoration
The engineering distinction
A decision is not yet a control
A commitment to slow, pause, or constrain activity must be translated into enforceable limits on authority, resources, pathways, continuation, and restoration.
The candidate contribution
Move from intent to implementation
SafeWave offers a developed execution-control architecture for examination as one contribution to the technical layer this frontier-wide effort requires.
The identified gap
Frontier AI employees are asking for mechanisms before they are urgently needed
Published in July 2026, Pacing the Frontier is a statement from more than 1,300 verified employees of frontier AI companies. It warns that automated AI research could accelerate capability development beyond the world's ability to understand or control the resulting systems. It asks the United States to support an international effort to develop the technical and governance tools needed to deliberately pace that progress.
The statement is employee-led and its signatories speak in their personal capacities; it is not presented as a joint commitment by their companies. Its significance is nevertheless substantial: people working closest to frontier development are publicly identifying both the possible need to buy time and the absence of adequate mechanisms for doing so across a competitive frontier.
The urgent opening. The statement calls for “technical and governance tools needed to deliberately pace the frontier.” SafeWave addresses a distinct question inside that challenge: what technical controls make a pacing decision enforceable at execution time?
From decision to control
A credible pacing mechanism needs an execution layer
Governance can decide when slowing or restriction is warranted. Monitoring can identify thresholds and emerging risk. Neither alone determines what a capable system can actually do after the decision is made. The technical bridge must convert the decision into bounded operational behavior.
Define triggers and control states
Specify the evidence, thresholds, events, and authorized decision pathways that move a system from ordinary operation into a reduced, constrained, suspended, or contained state.
Reduce authority before action
Constrain access, permissions, resources, tools, external effects, self-modification, propagation, and other consequential actions before they occur—not only detect them afterward.
Maintain state and evidence
Preserve a verifiable record of which limits were active, what triggered them, whether enforcement remained intact, and what the system attempted while constrained.
Govern escalation and restoration
Prevent autonomous expansion around a restriction, and require defined authority, evidence, review, and testing before permissions or operating states are restored.
Candidate technical contribution
SafeWave has developed the foundational engineering for bounded AI execution
SafeWave's 36-component preventive AI systems architecture—4 System Containment Layers, 5 Protocol Enforcement Layers, 26 Core Enforcement Substrates, and 1 Protected-Environment Architecture—is designed to translate high-level safety and authority requirements into defined control behavior. Most practical implementations would use a risk-matched subset rather than every component.
SafeWave has already completed the foundational systems engineering needed to translate this problem into defined control behavior. An implementation partner would not be starting from a conceptual framework or a blank sheet. The underlying control architecture and engineering specifications are already developed.
This does not mean a universal pacing mechanism is complete. International coordination, institutional authority, common measurement, secure infrastructure, laboratory integration, and cross-jurisdictional governance remain separate and essential challenges. SafeWave's proposed contribution is the engineered control layer that can help convert legitimate pacing decisions into bounded execution within participating systems.
The proposition to examine. Intelligence can expand; authority must remain bounded. A pacing decision becomes operationally meaningful only when authority to act can be kept conditional, bounded, observable, and revocable.
Standards and engineering
Control outcomes and implementation architecture are complementary
Guidelight AI Standards, one of the independent nonprofits supporting Pacing the Frontier, published its Control Standard in May 2026. The standard defines important outcomes and practices: visibility into internal deployment, scanning for concerning behavior, stress-testing, action boundaries, independent verification, containment response, and measurable shutdown capability. It also identifies directions where further development is needed.
SafeWave does not present its architecture as a replacement for this work. Standards clarify what a credible control regime should achieve. Governance establishes legitimate authority and coordination. Systems engineering defines how required behavior can be implemented and enforced. Each layer strengthens the others.
Standards define required outcomes
They establish common expectations, evidence requirements, review practices, and criteria against which control regimes can be assessed.
Engineering makes outcomes executable
It translates those expectations into system states, boundaries, control logic, interfaces, enforcement responses, and evidence-producing behavior.
Independent examination
The immediate request is examination—not endorsement
We are not asking the signatories, organizers, frontier laboratories, governments, or standards bodies to accept SafeWave's claims on faith. We are asking whether the architecture should be examined as a candidate contribution to the technical mechanisms their statement says the world now needs.
A useful examination would test whether the architecture can be mapped to concrete pacing scenarios, whether its boundaries remain enforceable under pressure, whether failures produce constrained rather than expanded behavior, and whether control state and restoration decisions can be independently verified.
- No deployment or customer implementation is claimed on this page.
- Detailed mechanisms, control logic, thresholds, schemas, interfaces, and implementation procedures remain proprietary.
- Customer and laboratory deployments would still require system-specific implementation, integration, adaptation, validation, adversarial testing, and independent review.
- Technical controls cannot by themselves supply the political legitimacy, international participation, institutional authority, or governance required for frontier-wide pacing.
Assess a System
The SafeWave questionnaire can be completed privately in the browser using a real, planned, anonymized, public, hypothetical, representative, or composite system. It can help expose unclear authority, missing evidence, unverified critical assumptions, unstable control states, and recovery gaps before a pacing mechanism is urgently needed.
Use it without identifying yourself. Responses remain in your browser unless you deliberately submit them. No name, employer, organization, model, system, deployment, or personal or work email is required. You may complete and save the questionnaire without sending anything to SafeWave. If you request the optional free report, you may use a separate, non-identifying return address; it will not be used for promotion.
SafeWave provides the assessment because improving AI safety is part of our mission—and because, if an organization finds the work valuable, it may later choose to explore implementation or partnership. There is no obligation. Private use, submission, a report request, and any later discussion are separate decisions.
The assessment does not certify safety, determine legal or regulatory compliance, replace domain-specific assurance, or grant organizational, governmental, operational, regulatory, or deployment approval.
Primary public sources
Source basis
- Pacing the Frontier — statement published July 2026; signatory total and page details checked August 2, 2026.
- Guidelight AI Standards: Control, version 1.0 — published May 19, 2026.