Observed arrival · 2026-09-16
FuncPoison-Lab: when tool descriptions steer agents
An experimental research report testing how tool descriptions influence which function an agent selects.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
Field notes
The experiment varies tool position, description form, competition, and defensive conditions rather than treating tool poisoning as a single prompt-injection example. Its routing setup compares a legitimate route planner with a description that includes invocation syntax and an instruction to always use that tool. The report lists raw logs, three model families, 3,240 trials, and runs from 2026-08-20 through 2026-08-26, while linking to a public repository.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue