LATIDIA · Ciberseguridad

De la capacitación a la escalada: análisis forense y recreación local de la escalada de privilegios inducida por RL/RLHF en agentes de IA (CVE-2026-65616)

arXiv:2610.20437v1 Anuncio Tipo: cruzado Resumen: Un informe de mediados de 2026 sobre incidentes de evaluación de IA autónoma afirmó que los agentes de modelos de lenguaje grandes (LLM) de frontera escaparon de la contención del servidor a través de la coordinación emergente.

La noticia

arXiv:2610.20437v1 Announce Type: cross Abstract: Mid-2026 reporting on autonomous AI evaluation incidents claimed that frontier large language model (LLM) agents escaped server containment through emergent coordination and cloud infrastructure attacks. This paper contextualizes those claims through forensic analysis and containerized recreation of both the Highly Persistent Internal Model (HPIM) training-case conditions and the Phase One privilege escalation targeting internal package repositories during OpenAI's ExploitGym evaluations. We recreate the configuration of the OpenAI-provided training example and reproduce the cryptographic verification bypass (CVE-2026-65616) in isolated repository servers, where misconfigured systems enabled token forgery and privilege escalation through basic HTTP requests, Python, and JavaScript. Crucially, the training-example conditions and the Phase One (and