acceptodds
Under review as a conference paper at ICLR 2027

Perceptive Closure: Identifying Authorization-Resource Counterfactuals

Abstract

Resource-specific claims about governed authorization—for example that representation is unnecessary, substitutable, or structurally required—are themselves counterfactual claims. We ask when such claims are identified. For a sealed authorization target, PERCEPTIVE CLOSURE (PC) represents evidence admission (), certificate representation (), and epistemic authority () as extensional contract coordinates and evaluates closure under every subset freeze. A complete contract yields a resource-freezing signature ; source/specification evidence yields the compatible identified set . We prove three stages. First, unrestricted controller behavior need not identify resource-freezing dependence: behaviorally equivalent systems can have arbitrary finite differences, or finite-versus-structural differences, under an freeze. Second, one fixed implementation description can admit source-consistent - and -typed completions with different freeze values. Third, conditional on explicitly declared semantic anchors and the source-exposed intervention boundary, touch is mechanically determined; interface-equivariant resource-support bisimulation is a sufficient certificate for invariance across interface-preserving refactorings. We then show that this identification boundary is operational rather than merely constructed. Apache Polaris is naturally boundary-underidentified, whereas a pre-existing XACML 3.0/AuthzForce 21.2.0 environment point-identifies an evidence-touching repair with T(q) = E, , . The XACML policy and native engine are unchanged and no manual action labels are used. PC-TAU supplies controlled nondegenerate freeze geometry, and live tool-use episodes establish finite-fallback realizability. PC does not claim prevalence, universal safety, causal decomposition inside mixed actions, or a universal ontology. Its claim is narrower: planning evaluates closure under a specified resource counterfactual; PC asks whether that resource counterfactual is warranted by the declared semantics in the first place.

Primary area:
alignment, fairness, safety, privacy, and societal considerations
Keywords:
counterfactual reasoning, partial identification, authorization, access control, semantic identification, AI governance, formal methods, Perceptive Closure

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_HfnvOpenReview ↗
Rating
2
Confidence
1
Informativeness
1
Presentation
1
Soundness
2
Read review
Paper summary

The paper presents a theoretical framework to study authorization governance using formal methods, where it particularly studies identifiability of certain counterfactual claims. It proposes a contract perceptive closure that represents evidence admission, certificate representationn and epistemic authority and evaluate E- and R-consistent semantic completion under freezing. Theoretical results are derived within this framework. The framework is applied to analyze Apache Polaris and XACML 3.0/AuthzForce 21.2.0.

Critical strengths
  • I'm not familiar with the jargons, notations, and presentations as they are presented. It is possible that the framework and theory presented here are of high value in certain specific domains.
Critical weaknesses
  • The paper is very dense and the notations are not properly defined when they are first used. For example section 2 start with a definition of the PC contract as a 15-component tuple. Barely anything about these 15 objects are defined except for their type signatures. There is no explanation about the interpretation of most of these objects, no definition of the central object SS or its interpretation. The three key objects, E,R,AE, R, A are not formally defined anywhere either.
  • The paper does not explain how its theoretical framework and consequences connect to any potential application in relation to its primary area "alignment, fairness, safety, privacy, and societal considerations". I believe the connection is to be drawn through authorization governance to AI agents, but not detail is given.
Questions for authors
  • Could you please carefully define the notations used in section 2 and their interpretations, or give references where to find them from literature?
  • Could you please explain why contributions 1 and 2 are nontrivial and not tautological?
  • Will implementation source code of PC be available?
Reviewer_rzSbOpenReview ↗
Rating
1
Confidence
2
Informativeness
1
Presentation
1
Soundness
3
Read review
Paper summary

I am afraid I do not feel sufficient expertise to understand this contribution.

The authors seem to make a distinction between a resource intervention and the source and evidence for it. However it seems as if the burden of this relation is largely then hidden away and abstracted and the positive identification should carry all the weight.

It could be that the distinction between available information and admissable evidence is important and indeed raised elsewhere. However I just do not see how the couterfactuals are to be identified.

Critical strengths

None.

Critical weaknesses

I just do not appreciate what is new here nor how significant it is and to what extent pushing the burden to the counterfactuals solves any of the concerns raised by the authors.

I could be missing things here due to lack of experties.

Questions for authors

As above.

Reviewer_BJjKOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary

The paper seems to address a problem in the context of authorization. However, I just cannot follow the paper at all. There is no clear problem statement. Specialized symbols, terms and phrases are used without proper introduction and definition. I almost gave up after reading the abstract and has to give up after page 2, as I cannot make any sense out of it.

Critical strengths

Given the significant presentation issue, I cannot think of any strengths of the paper. Maybe the paper makes good technical contributions, but at its current presentation, I don't think the paper will make any sense to most of the audience of ICLR.

Critical weaknesses

Significant presentation issues. Starting from the abstract, the paper dives into its own terminologies and special symbols without explaining the problem at all.

Questions for authors

None. The paper should be totally re-written with proper motivation, problem statement and introduction/definitions of terms and concepts used.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.