Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJan 30, 2026Canada

CoherenceGuard

Nadine Squires · Team QUANTARA

Submitted to The Technical AI Governance Challenge. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: CoherenceGuard

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

Current AI governance relies on tracking physical infrastructure (data centers, GPU clusters, power consumption), but this approach fails against architecturally efficient "dark" systems that achieve frontier capabilities with minimal detectable footprint. CoherenceGuard solves this enforcement gap. CoherenceGuard is a deterministic, open-source governance engine that audits AI systems at the structural level rather than the physical level, enabling enforceable international agreements even when traditional monitoring fails. It operates through three integrated layers:

1. Evidence Layer: Fixed-precision manifold mapping that generates bit-for-bit identical measurements across any hardware platform—eliminating "floating-point drift" that undermines legal defensibility.

2. Interpretation Layer: Swappable policy adapters that simultaneously benchmark systems against multiple standards (EU AI Act, G7 roadmaps, Apart Challenge protocols) using a single coherence-focused assessment.

3. Decision Layer: Automated enforcement triggers with immutable audit trails, providing the transparency required for whistleblower protections and international oversight.

This architecture operationalizes currently undefined "State-of-the-Art risk modeling" requirements, detects incremental capability gains that evade threshold-based alarms ("Boiling Frog" risk), and provides hardware-agnostic determinism for legally binding enforcement.

CoherenceGuard is fully prototyped with working code, tested on stealth architectures, and designed to provide the verification infrastructure for the 2027 binding agreement roadmap—directly advancing the challenge's goals of creating practical verification, compliance, and coordination tools for frontier AI safety.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for AI safety if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical AI safety problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. I’m not in a position to evaluate the technical claims about ASIOS and CoherenceGuard, but I worry that you claim to be open sourcing very powerful capabilities which would have deleterious effects on the prospects for effective governance of frontier AI. You are relying on "International Binding Agreements" to enforce use, but your own technology (ASIOS) makes the physical detection of violators (via power/heat/compute) impossible. What prevents an actor from using ASIOS to build hidden AI on consumer hardware while simply deleting the CoherenceGuard auditing modules?

    I guess that you are operating from an assumption that widely proliferated, highly capable AI systems will lead to desirable outcomes. This probably in turn rests on the idea that the dangers of misused or misaligned AI can be addressed “head on” through direct competition by others who have aligned AI and pursue widely beneficial aims, like defense. I think this is likely misguided and it relies on whether there will be methods of AI misuse which are highly offense-favored, in that their negative effects are difficult to defend against.

    Read full reviewShow less

  2. The submission is very difficult to evaluate because the core claims are buried under dense, nonstandard jargon (e.g., "recursive geometric coherence," "paradox-preserving intelligence modeling," "entropy-aware degradation mechanics") that is not grounded in established literature. When checking the linked repositories, the technical claims are not reflected in the actual code. I'd suggest using standard terminology, grounding claims in existing work, and building a small working prototype that demonstrates one concrete capability.

Cite this project

@misc{squires2026coherenceguard,
  title = {{CoherenceGuard}},
  author = {Nadine Squires},
  year = {2026},
  month = jan,
  note = {Submitted to The Technical AI Governance Challenge, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/coherenceguard-1fht}},
  url = {https://apartresearch.com/sprints/projects/coherenceguard-1fht}
}

More from The Technical AI Governance Challenge

All 47 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026