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.

Predictable and Performant Reactive Synthesis Modulo Theories via Functional Synthesis

Reactive synthesis is the process of generating correct controllers from temporal logic specifications. Classical LTL reactive synthesis handles (propositional) LTL as a specification language. Boolean abstractions allow reducing LTLt specifications (i.e., LTL with propositions replaced by literals from a theory calT), into equi-realizable LTL specifications. In this paper we extend these results into a full static synthesis procedure. The synthesized system receives from the environment valuations of variables from a rich theory calT and outputs valuations of system variables from calT. We use the abstraction method to synthesize a reactive Boolean controller from the LTL specification, and we combine it with functional synthesis to obtain a static controller for the original LTLt specification. We also show that our method allows responses in the sense that the controller can optimize its outputs in order to e.g., always provide the smallest safe values. This is the first full static synthesis method for LTLt, which is a deterministic program (hence predictable and efficient).