AI
SEAD: A State-Based Perspective on Attack and Defense in Tool-Using Agents
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 30/09/2026, 10:01:17 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 09:18:24 UTC.
Briefings
SEAD proposes a state-based framework for analyzing tool-using agents, distinguishing between visible interactions and underlying states. The DART system decomposes harmful goals into locally plausible steps using feedback from actual tool execution, while SAGE investigates relevant state via read-only queries before allowing or blocking actions. Across four target models, DART improved semantic attack success by 18.8 to 35.9 percentage points over a baseline, with SAGE preserving 95.79% of benign trajectories and intercepting 92.73% of harmful paths. In online evaluation, SAGE reduced DART's executable attack success from 48.0% to 4.0%.
Key points
- DART decomposes harmful goals into locally plausible steps using feedback from actual tool execution to guide trajectory search.
- SAGE investigates relevant state through read-only queries before allowing or blocking each action, including those proposed after a block.
Context and limitations
Results are based on four target models and online attack-defense evaluation; performance gains may vary across different environments and attack methods.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/09/2026, 10:01:17 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeig7pacvvcjcwxvvfb47zqlapgylbu6jth3oygfsnnuoq7tdkmpfra