Multimodal-SEA: Benchmarking AI Agent Security under Multimodal Social Engineering Attacks
Factual evidence
What the source reports
Researchers introduced Multimodal-SEA, a benchmark testing AI agent security against multimodal social engineering fraud scenarios.
Inspect the evidence
- Inclusion basis
- Enterprise AI
- Publisher and source type
- arXiv cs.CL — Computation and Language · RESEARCH
- Published by source
- 6 November 2026
- Collected by OneBench
- 7 Nov 2026, 02:01 UK
Stored source excerpt
arXiv:2611.08281v1 Announce Type: new Abstract: Personal AI agents face social engineering attacks supported by forged multimodal content, but existing evaluations mainly focus on text-based dialogue,…
Short excerpt from the collected text, not the full source. Use the source link to read it in context.
The factual summary is a OneBench synthesis, not a quotation or independent verification. Collection time is not publication time. Open the source for its full context; related reporting can share the same underlying announcement.
OneBench interpretation
Institutional assessment
So what
Multimodal AI agents remain vulnerable to complex visual and audio social engineering tactics derived from real-world fraud scenarios.
Do what
Review agent security evaluation criteria with model risk teams testing multimodal customer service and authentication workflows.