AI Security: Bunker, Hardening, and Safety Builds

This page collects the NosisTech security work in three areas. Bunker is an offline scanner for crypto addresses. The hardening notes cover this website and a home AI lab. The safety builds test controls around AI models.

Every article records real configuration and real results. The safety builds are small educational demos, so each one shows a single control clearly.

Bunker: The Full Series

Bunker is a free, open-source scanner that shows which crypto addresses expose their public key. It runs only on your device and covers more than 100 networks.

Open Bunker, or read the 13 articles in order.

#ArticleWhat it covers
1Which of Your Crypto Addresses Show Their Public Key?What Bunker checks, and the plan it gives.
2The Sealed Box: When a Crypto Address Shows Its Public KeyWhat makes a key visible, and the seven results.
3How Each Network Shows the Public KeyBitcoin, Ethereum, Solana and more, with real Bunker scans.
4Inside the Five-Step Engine of BunkerThe parser, 17 ordered detectors, and the rules that build the plan.
5How Bunker Catches Typos in Crypto AddressesChecksums in Bitcoin, Ethereum, XRP and Algorand addresses.
6How Bunker Blocks 19 Kinds of SecretsThe guard that stops seed phrases and private keys before any address is read.
7A Web Page That Cannot Send Your DataA strict Content Security Policy, Trusted Types, and no storage.
8Zero Dependencies in BunkerNo npm packages, CI with no third-party actions, and a reproducible build.
9The Hard Cases in BunkerTaproot, Zcash, Canton, and shared address formats.
10Bunker for AI Agents: A Local CLI and MCP ServerRead-only tools for agents, with no web API.
11Designing Bunker for Phones, Screen Readers and Two LanguagesA WCAG 2.2 AA audit with 0 violations.
12The Bunker Build Log46 problems fixed and 30 decisions recorded across seven phases.
13Bunker Against the OWASP MCP Top 10The MCP server checked item by item, and three limits that no tool can fix.

Securing This Website

Three articles on how nosistech.com is protected, with the real rules and settings.

#ArticleWhat it covers
1Blocking AI Scrapers, Bots, and Brute ForceRules in robots.txt for AI crawlers, WAF rules in Cloudflare, and login protection.
2Cloudflare Cache Rules, WAF, and the REST API LeakOne WAF rule that blocks the WordPress users endpoint.
3Fixing a WordPress Admin Bar Leak in the CacheThe purge order, and a least-privilege Cloudflare token.

Securing a Self-Hosted AI Lab

ArticleWhat it covers
Securing a Self-Hosted AI Lab with Cloudflare Zero TrustAutomated probes arrived within hours, and Zero Trust Access now checks every visitor.

The lab setup itself is in the Local AI hub.

AI Safety Builds

These builds come from the agent patterns library. Each one adds a control around an AI model.

ArticleWhat it covers
Prompt Injection Defense AgentFlags suspicious text before it reaches the main model.
LLM Guard: Building an AI Safety GatewayScans inputs, redacts data, and checks outputs through LiteLLM.
Promptfoo Redteam Eval AgentTests AI answers with visible assertions and short verdicts.
Hallucination Detection AgentChecks answers against source context and flags unsupported claims.
PentestGPT Authorized Review AgentPlans scope, review, and evidence for authorized testing.
OpenHands Sandbox PlannerPlans software tasks behind a review gate.
Aegis Constraint AgentFixed rules make the decision, and the model only explains it.
Self-Improving Agent With GuardrailsProposes prompt changes when metrics drop, and sends risky ones for approval.

More From NosisTech

PageWhat it covers
Local AIPrivate models on laptops, home labs, and VPS servers.
AI GovernanceFree policy templates and governance builds.
BuildsBunker, Korvin, and the other bodies of work.
All articlesEvery NosisTech article, newest first.