QuillGuard

Since the initial storm of AI agents in early 2024, the Web3 market has seen immense growth. This sector’s flagship token $VIRTUAL, already boasts a market cap of $1.12 billion, marking a five‑fold jump since Q4 2024 and becoming the largest among “ai agent” projects. The wider AI‑agent token basket has ballooned from near‑zero to $15 billion, and multiple trackers now project it could surge past the $200 billion mark by the end of  2025 as new launches proliferate.

A large part behind the success of these agents lies in their ability to analyse market data, and perform complex onchain operations. However the same abilities expose them to grave dangers. Onchain agents in particular, make enticing targets for adversaries as they often deal with huge capital flow. All such problems stem from the use of LLMs as the primary decision making engine, which fail to effectively enforce prompt-level runtime policies. We propose QuillGuard: Adversarial Dynamic AgentRails, a self evolving-adversarial system which can autonomously detect vulnerabilities across known attack surfaces, and dynamically implement guardrails against them to safeguard them from the same attacks. We also theorise how such a solution, while being highly effective for Web3 agents, can be universally applied to any agent