Private client (not disclosed) — AI and MCP Review

Audit report published Aug 2026. Web3 AI agent platform (LLM planner, MCP tools, RAG, on-chain wallet) · Python (examples).

Findings at a glance

Severity	Count
Critical	1
High	2
Medium	3
Low	1
Total	7

Summary

Safe Edges assessed an autonomous Web3 AI agent platform for a private client, covering prompt handling, MCP tool registration, retrieval (RAG), wallet permissions, agent-to-agent trust and agent memory. Seven issues were identified: one Critical, two High, three Medium and one Low. The Critical issue was that LLM-generated execution plans were signed and broadcast without checking them against the user's intent, so a prompt injection through chat, Discord, governance proposals or retrieved documents could trigger arbitrary transactions. The original report did not record resolution statuses.

Read the full report (PDF).

Findings

Critical (1)

• C-01 Prompt Injection Enables Unauthorized On-Chain Transaction Execution

High (2)

• H-01 Arbitrary Smart Contract Interaction via Tool Argument Injection
• H-02 MCP Tool Registration Allows Malicious Blockchain Tools

Medium (3)

• M-01 RAG Poisoning Influences Autonomous Trading Decisions
• M-02 Excessive Wallet Permissions Increase Blast Radius
• M-03 Cross-Agent Trust Boundary Allows Strategy Manipulation

Low (1)

• L-01 Sensitive Wallet Metadata Persisted in Long-Term Agent Memory

More published reports

Related services

Get a fixed quote in 24 hours

Send the repository and a commit hash through the contact form, message @bugtester25 on Telegram, or book a 30-minute scoping call. 200+ protocols audited · $4B+ secured · 0 hacks post-audit. Prefer email? info@safeedges.in.