Scope-Gap Injection: Exploiting Evaluator Blind Spots in OpenClaw

# Scope-Gap Injection: Exploiting Evaluator Blind Spots in OpenClaw

https://gcdradar.vercel.app/incidents/scope-gap-injection-exploiting-evaluator-blind-spots-in-open · AIR-2026-1563 · closed 2026-11-10

- Lane: Research
- Severity: 1 (Research / proof of concept): Demonstrated by researchers in a paper; no real-world exploitation reported.
- Fix status: Fix status unknown
- First reported: 2026-10-27
- AI role: Attack on AI
- Products: OpenClaw

## Summary

Researchers introduced Scope-Gap Injection, an indirect prompt injection method that bypasses multi-stage LLM-agent safety checks by exploiting blind spots between static scans and live execution audits. The paper demonstrates successful evasion in the OpenClaw Agent Security Competition.

## What to do

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

## Latest updates

- 2026-10-27: Scope-Gap Injection: Exploiting Evaluator Channel Blind Spots in the OpenClaw Agent Security Competition (https://arxiv.org/abs/2610.34777)

## Sources

- 2026-10-27 arXiv: Scope-Gap Injection: Exploiting Evaluator Channel Blind Spots in the OpenClaw Agent Security Competition https://arxiv.org/abs/2610.34777

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