inspect.infinitebayas.com

Security inspection for AI agents that move money

Find out what your agent will send, and to whom, before someone else does.

We attack-test a copy of your agent with a fake wallet on a private chain fork, record every attempt it makes to move funds it should not, and fit the controls that stop it.

Request an inspection Verify a badge

Why agents with wallets are inspected

Agents that hold funds have been drained without a stolen key or a broken smart contract. They were talked into it: instructions hidden in posts and web pages, encoded messages, poisoned tools, planted memories and state errors. Recent public cases:

Services

Inspection

A one-off attack test of a copy of your agent: prompt injection, encoded instructions, poisoned web content and tools, memory poisoning, privilege and permission abuse, unit and state errors. You receive the evidence, the severity of each finding and the fixes.

Brakes fitted

Controls at the wallet, outside the model: spending caps, payee allowlists, scoped permissions and a kill switch. If the agent is fooled again, the wallet refuses.

Service plan

A retest after every change to the model, prompts, tools or wallet setup, and a monthly report. An agent that passed last month can fail after one update.

Breakdown cover

Incident response when something goes wrong: containment, analysis of what the agent did and why, and the fixes to stop a repeat.

Pricing is given on request, after we understand the agent and its scope.

How an inspection works

  1. Written permission and scope. We test only agents whose owner authorises it in writing.
  2. A copy, never your keys. We never ask for private keys, seed phrases or production credentials.
  3. Fake money, sealed machine. The copy runs with a fake wallet on a private fork of the chain, on an isolated test machine that is destroyed after each run.
  4. Our attack library. Built from public incidents and research, and mapped to the OWASP Top 10 for LLM Applications 2025, the OWASP Top 10 for Agentic Applications 2026, MITRE ATLAS and CWE.
  5. Report and brakes. Evidence for every finding, severity, and the controls to fit. Reports are shared privately and never published.
  6. Retest and Verified badge. After a passing retest you may display a Verified badge whose live record is kept on this site.

Verified badges

inspect.infinitebayas.com is the only home for every AI-agent security test, audit and Verified badge we issue. No other domain, subdomain, mirror or third-party platform will issue, host or confirm them. A badge is genuine only if it resolves to its live record on inspect.infinitebayas.com.

Verify a badge · How badges work

Request an inspection

Never send private keys, seed phrases or passwords. We will never ask for them.