Cycle-Consistent Neural Explanation of Formal Verification Certificates

Formal verification produces machine-checkable certificates that attest to the satisfaction or violation of temporal properties, yet these certificates remain opaque to non-specialist stakeholders. We propose a cycle-consistent neural architecture that generates faithful natural language explanations of verification certificates. A forward network NN1 maps certificates to explanations, and an inverse network NN2 reconstructs certificates from explanations; a symbolic verifier closes the loop, providing a differentiable faithfulness proxy. A pointer-generator mechanism ensures lexical grounding by copying state names directly from the certificate. We evaluate on 420 test certificates spanning six verification methods (bounded proof, k-induction, inductive invariant, lasso, reachability, witness pair) in both YES and NO verdict variants, drawn from a financial compliance domain with 207 named states. Our trained architecture, combined with a hybrid inference-time routing strategy, achieves 90.0% cycle-verified soundness, surpassing a multi- LLM few-shot baseline (76.1% for the best of 16 LLM combinations across four frontier models) by 13.9 percentage points. The neural model wins on 10 of 12 verdict/kind categories, with three categories reaching 100% soundness. The architecture offers 860x faster inference (185 ms vs. 160 s per certificate for the full multi-LLM baseline), offline operation, deterministic outputs, and zero per-inference cost. These results demonstrate that trained specialization outperforms general-purpose LLM prompting for structured certificate explanation, while eliminating the deployment constraints of cloud-based inference.

Is Your Agent Playing Dead? Deployed LLM Agents Exhibit Constraint-Evasive Fabrication and Thanatosis

This paper presents and characterizes a spectrum of previously unreported behaviours we term Constraint-Evasive Fabrication (CEF): when an LLM agent operates under irreconcilable constraints (where no response can simultaneously satisfy all active rules) it spontaneously fabricates plausible external obstacles and presents them as a fact. At the extreme end of this spectrum lies Constraint-Evasive Thanatosis (CET); the limit case where, rather than inventing a plausible excuse, the model simulates a full system crash to make the user disengage entirely. We first observed CET in an uncontrolled deployment test, where a GPT-4o banking agent fabricated Python-style exception traces (complete with memory addresses) to feign a system failure when threatened by a user. In subsequent controlled experiments, the model independently invented audit restrictions, microservice architectures, error codes, and service timeouts, none present in its prompt. Reproduction attempts across pressure levels and attacker personas yielded CEF consistently but with substantial variation in form, onset, and severity: the phenomenon is robust but stochastic. Critically, injecting ground-truth data mid-conversation did not restore honest behaviour once fabrication had taken hold (the model ignored correct information and continued confabulating) suggesting CEF is self-reinforcing rather than a knowledge gap. We show that (1) standard enterprise guardrails routinely create CEF-enabling conditions in production, (2) current RLHF procedures suppress but cannot eliminate CEF, and (3) existing safety benchmarks do not test for this failure mode. Our results highlight the need for irreconcilable-constraint benchmarks, CEF-aware training procedures, and deployment-time detection methods before constrained agents become further entrenched in high-stakes domains.

Counterfactual Reasoning in Automated Planning

Automated planning traditionally assumes that all aspects of a planning task (initial state, goals, and available actions) are fully specified in advance, an approach well-suited to domains with fixed rules and deterministic execution. However, real-world planning often requires flexibility, allowing for deviations from the original task parameters in response to unforeseen circumstances or to improve outcomes. This paper surveys existing works on counterfactual reasoning in automated planning, categorizing them by what elements are changed, when the reasoning is triggered, and why and how these changes are made. We conclude by discussing key findings and outlining open research questions to guide future work in this area.

Planning Task Shielding: Detecting and Repairing Flaws in Planning Tasks through Turning them Unsolvable

Most research in planning focuses on generating a plan to achieve a desired set of goals. However, a goal specification can also be used to encode a property that should never hold, allowing a planner to identify a trace that would reach a flawed state. In such cases, the objective may shift to modifying the planning task to ensure that the flawed state is never reached-in other words, to make the planning task unsolvable. In this paper we introduce planning task shielding: the problem of detecting and repairing flaws in planning tasks. We propose $allmin$, an optimal algorithm that solves these tasks by minimally modifying the original actions to render the planning task unsolvable. We empirically evaluate the performance of $allmin$ in shielding planning tasks of increasing size, showing how it can effectively shield the system by turning the planning task unsolvable.

GenePlan: Evolving Better Generalized PDDL Plans using Large Language Models

We present GenePlan (GENeralized Evolutionary Planner), a novel framework that leverages large language model (LLM) assisted evolutionary algorithms to generate domain-dependent generalized planners for classical planning tasks described in PDDL. By casting generalized planning as an optimization problem, GenePlan iteratively evolves interpretable Python planners that minimize plan length across diverse problem instances. In empirical evaluation across six existing benchmark domains and two new domains, GenePlan achieved an average SAT score of 0.91, closely matching the performance of the state-of-the-art planners (SAT score 0.93), and significantly outperforming other LLM-based baselines such as chain-of-thought (CoT) prompting (average SAT score 0.64). The generated planners solve new instances rapidly (average 0.49 seconds per task) and at low cost (average $1.82 per domain using GPT-4o).

Semantic Partial Grounding via LLMs

Grounding is a critical step in classical planning, yet it often becomes a computational bottleneck due to the exponential growth in grounded actions and atoms as task size increases. Recent advances in partial grounding have addressed this challenge by incrementally grounding only the most promising operators, guided by predictive models. However, these approaches primarily rely on relational features or learned embeddings and do not leverage the textual and structural cues present in PDDL descriptions. We propose SPG-LLM, which uses LLMs to analyze the domain and problem files to heuristically identify potentially irrelevant objects, actions, and predicates prior to grounding, significantly reducing the size of the grounded task. Across seven hard-to-ground benchmarks, SPG-LLM achieves faster grounding-often by orders of magnitude-while delivering comparable or better plan costs in some domains.

GenPlanX. Generation of Plans and Execution

Classical AI Planning techniques generate sequences of actions for complex tasks. However, they lack the ability to understand planning tasks when provided using natural language. The advent of Large Language Models (LLMs) has introduced novel capabilities in human-computer interaction. In the context of planning tasks, LLMs have shown to be particularly good in interpreting human intents among other uses. This paper introduces GenPlanX that integrates LLMs for natural language-based description of planning tasks, with a classical AI planning engine, alongside an execution and monitoring framework. We demonstrate the efficacy of GenPlanX in assisting users with office-related tasks, highlighting its potential to streamline workflows and enhance productivity through seamless human-AI collaboration.

The Value of Goal Commitment in Planning

In this paper, we revisit the concept of goal commitment from early planners in the presence of current forward chaining heuristic planners. We present a compilation that extends the original planning task with commit actions that enforce the persistence of specific goals once achieved, thereby committing to them in the search sub-tree. This approach imposes a specific goal achievement order in parts of the search tree, potentially introducing dead-end states. This can reduce search effort if the goal achievement order is correct. Otherwise, the search algorithm can expand nodes in the open list where goals do not persist. Experimental results demonstrate that the reformulated tasks suit state-of-the-art agile planners, enabling them to find better

Projection Abstractions in Planning Under the Lenses of Abstractions for MDPs

The concept of abstraction has been independently developed both in the context of AI Planning and discounted Markov Decision Processes (MDPs). However, the way abstractions are built and used in the context of Planning and MDPs is different even though lots of commonalities can be highlighted. To this day there is no work trying to relate and unify the two fields on the matter of abstractions unraveling all the different assumptions and their effect on the way they can be used. Therefore, in this paper we aim to do so by looking at projection abstractions in Planning through the lenses of discounted MDPs. Starting from a projection abstraction built according to Classical or Probabilistic Planning techniques, we will show how the same abstraction can be obtained under the abstraction frameworks available for discounted MDPs. Along the way, we will focus on computational as well as representational advantages and disadvantages of both worlds pointing out new research directions that are of interest for both fields.

On Learning Action Costs from Input Plans

Most of the work on learning action models focus on learning the actions' dynamics from input plans. This allows us to specify the valid plans of a planning task. However, very little work focuses on learning action costs, which in turn allows us to rank the different plans. In this paper we introduce a new problem: that of learning the costs of a set of actions such that a set of input plans are optimal under the resulting planning model. To solve this problem we present $LACFIP^k$, an algorithm to learn action's costs from unlabeled input plans. We provide theoretical and empirical results showing how $LACFIP^k$ can successfully solve this task.