Spoofing the Latent State: Imperceptible Goal-Image Attacks on LeWM

# Spoofing the Latent State: Imperceptible Goal-Image Attacks on LeWM

https://gcdradar.vercel.app/incidents/spoofing-the-latent-state-imperceptible-goal-image-attacks-o · AIR-2026-1740 · live

- Lane: Research
- Severity: 1 (Research / proof of concept): Demonstrated by researchers with no known real-world exploitation.
- Fix status: Fix status unknown
- First reported: 2026-11-03
- AI role: Attack on AI
- Products: LeWorldModel

## Summary

Researchers demonstrated a test-time attack that derails planning in LeWorldModel by manipulating goal-image latents. The work shows that imperceptible perturbations can force the visual world model to misdirect control tasks.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-11-03: Spoofing the Latent State: Imperceptible Goal-Image Attacks Derail Planning in LeWorldModel (https://arxiv.org/abs/2611.00427)

## Sources

- 2026-11-03 arXiv: Spoofing the Latent State: Imperceptible Goal-Image Attacks Derail Planning in LeWorldModel https://arxiv.org/abs/2611.00427

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.