LATIDIA · Ciberseguridad

Aplicaciones de la IA de caja negra

arXiv:2610.32092v1 Tipo de anuncio: nuevo Resumen: Verificar la procedencia de los agentes LLM autónomos es un desafío de seguridad crítico. La atribución semántica existente se ofusca fácilmente por las indicaciones del sistema, mientras que el agua activa

La noticia

arXiv:2610.32092v1 Announce Type: new Abstract: Verifying the provenance of autonomous LLM agents is a critical security challenge. Existing semantic attribution is easily obfuscated by system prompts, while active watermarking conflicts with schema-constrained tool generation. We propose AgentTrace, an active framework that identifies the foundation model behind a black-box agent via observable execution traces. AgentTrace injects structurally ambiguous probes to trigger tokenization-sensitive serialization failures at rigid tool interfaces. While these initial failures provide coarse-grained separation across model families, the temporal sequence of subsequent repair operations reveals fine-grained, model-specific recovery behaviors. By modeling these multi-step trajectories with lightweight Markov chains, AgentTrace achieves 96.0% attribution accuracy under adversarial identity spoofing. Crucially, sequential repair