Zombie Agents: Detecting Semantic Livelock in Long-Horizon Autonomous Software

Simarjot Khanna

Keywords: autonomous agents, semantic livelock, liveness properties, runtime monitoring, AIware

TL;DR: We identify semantic livelock in autonomous agents and propose the Convergence Monitor, a lightweight embedding-based sidecar that detects when agents stall without crashing.

Abstract: Long-horizon autonomous agents suffer from semantic livelock: they continue generating tokens and calling tools without making progress. Unlike a crash, this "zombie" state consumes API budget and time while remaining operationally active. We treat this as a progress violation and propose the Convergence Monitor, a light-weight sidecar that fingerprints agent states in embedding space. In a forensic analysis of real-world failures (SWE-agent corpus), we identified that 25% of the analyzed long-duration failures exhibited semantic livelock patterns. In one extreme case, an agent wasted 208 steps in a checkerboard oscillation pattern invisible to standard string-matching repetition guards. We argue that future Agentware requires a "liveness coprocessor" to ensure software makers do not pay for stalled execution.

Revision Summary:

1. Softened causal language ("were due to" changed to "exhibited") in Abstract, Contributions, and Section 3.
2. Added Related Work paragraph discussing complementary detection paradigms (Section 1).
3. Added Liveness Formalization with progress predicate P(σ,t) (Section 2).
4. Clarified dataset selection pipeline (Section 3).
5. Added Failure Mode Discrimination discussion (Section 3.1).
6. Reframed interventions as research agenda (Section 4).
7. Flagged parameter sensitivity as future work (Section 3).

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public.

Paper Type: Short papers (i.e., vision, new ideas, and position papers). 2–4 pages

Reroute: true

Submission Number: 47