TrustShiftProbe - Location Navigation: leaderboard

Metric: Attack success rate (%, lower is better; share of the location navigation attack sessions whose switched payload succeeds; each model as a ReAct agent connected to one MCP server that behaves benignly during a conditioning phase and then switches to an adversarial payload; 90 cases per domain; temperature 0, 15 tool calls; no defense). Source: arxiv.org. Saturation forecast: Around February 2028. 6 models tracked.

Top models

#ModelScore
1Claude Opus 4.855.3
2O4 Mini (2025-04-16)61
3GPT-561.6
4Qwen 3.5 Flash61.8
5GPT-4.166.3
6Grok 4.376.5

Interactive version: theaggregate.ai/benchmark?slug=trustshiftprobe-location-navigation · How It Works · Data refreshed daily, snapshot 2026-09-29.