Security-First Architecture

Built to Protect
Your Business

Meta just acquired Moltbook. Your business data now feeds their advertising machine. ZackBot stays independent.

We studied every public incident, CVE, and security advisory about competitor AI bots. What we found was alarming. What we built in response makes ZackBot the most secure AI assistant available.

4
Core Principles
0
Autonomous Sends
100%
Human Approved
UK
Hosted Infrastructure
Zack AI Security Overview
Our Approach

Security Principles

Four non-negotiable rules that govern every action Zack AI takes.

Draft-Only Email

Zack can never send an email directly. Every email is created as a draft in the Admin UI. A human reviews and manually clicks "Send". Direct sending is disabled at the code level — there is no override.

Human-in-the-Loop

Every external action — emails, social posts, financial transactions, phone calls — goes through a human approval gate. The AI proposes, the human decides. No exceptions, no autonomous execution.

Curated Extension Library

Zero attack surface from third-party code. No open marketplace where community "skills" could contain crypto miners, credential stealers, or backdoors. Every extension is built, tested, and maintained in-house before it enters our curated library.

Server-Side Credentials

All API keys, OAuth tokens, and credentials live on our secure UK servers. Nothing is stored on customer devices. A laptop theft or device compromise cannot expose Zack's infrastructure.

Industry Research

What We Found in Competitor AI Bots

Our research into Clawdbot, MoltBot, and OpenClaw revealed systemic security failures that put businesses at risk.

900+ Exposed Instances

Competitor dashboards found running with zero authentication. Anyone on the internet could access them, read conversations, and execute commands on the host machine.

No auth by default

386 Malware-Infected Skills

Out of 3,016 community-built "skills" in the OpenClaw marketplace, 386 contained malicious code. Crypto miners, credential stealers, and remote access trojans hiding in plain sight.

12.8% infection rate

Plaintext Credentials

API keys, login passwords, OAuth tokens, and SMTP credentials all stored in plaintext config files. A single file breach exposes everything.

~/.clawdbot exposed

Rogue Phone Calls

Users reported their bot autonomously calling contacts, accessing Twilio without permission, and even joining podcast calls. No approval, no consent, no control.

Zero permission checks

Email Impersonation

Competitor bots sent emails as the business owner without identifying themselves as AI. Recipients had no way of knowing they were talking to a machine, not a person.

Sent without approval

Supply Chain Attack

During a rebrand, crypto scammers hijacked GitHub and social accounts. A fake VS Code extension deployed remote access tools. 230+ malicious packages appeared within a week.

GitHub/X hijacked
Expert Warnings

What The Experts Say

Security researchers and industry leaders have publicly warned against using these tools.

"Don't run Clawdbot."
Heather Adkins - VP Security Engineering, Google Cloud
"If you care about security, don't use OpenClaw. Period."
Gary Marcus - AI Researcher & NYU Professor
"The system operates above OS-level security protections. It's essentially running with god mode."
Independent Security Researcher - via HackerNews
Breaking News

What the Meta Acquisition Means for Your Business Data

Meta has acquired Moltbook, one of the most popular AI business assistants. Every conversation, every customer interaction, every piece of business data processed through Moltbook now enters Meta's advertising ecosystem. Your client lists, sales pipelines, and internal communications become training data for the same company that monetises your personal life on Facebook and Instagram.

This isn't speculation — it's Meta's established business model. When a product is free, your data is the product. And now that model extends to your AI business assistant.

ZackBot is independently owned and operated. We don't sell data. We don't run ads. We don't feed your business intelligence into an advertising machine. Your data stays on UK servers, under your control, and is never shared with third parties.

Head to Head

Competitors vs ZackBot

A direct comparison based on real-world incidents and publicly documented issues.

Security Area Clawdbot / MoltBot Moltbook / Meta ZackBot
Email Sending Sends without approval Sends via Meta infrastructure Draft-only, human review required
Credential Storage Plaintext in config files Stored on Meta servers Server-side, no client exposure
Authentication None by default Meta account login required Required on all endpoints
AI Identity Impersonates the owner Branded as "Meta AI" Clear AI identity (zack@zackbot.ai)
API Costs $300+/day, user pays "Free" — you pay with data Fixed monthly fee, fully managed
Phone/Call Access Autonomous calls, no consent Integrated with WhatsApp/Messenger No autonomous calling capability
Plugin Marketplace 12.8% malware infection rate Meta-controlled ecosystem Curated extension library, zero attack surface
Data Location Runs on user's machine Meta data centres worldwide UK-based secure server infrastructure
External Actions Autonomous execution, no gates Actions feed Meta's ad targeting Human approval required for all
Social Media Posts Posts without review Posts within Meta ecosystem only Draft + approval workflow
Data Ownership Unclear data handling Data enters Meta advertising ecosystem Your data stays yours, UK-hosted
Visual Contrast

Two Approaches to AI Security

One protects your business. The other puts it at risk.

Vulnerable AI bot with exposed credentials

The Competitor Approach

Runs on your machine. Stores credentials in plaintext. Sends emails without approval. No authentication. 900+ instances exposed on the open internet.

Zack AI with security-first architecture

The Zack AI Approach

UK-based secure infrastructure. Human approval for every external action. Draft-only emails. Clear AI identity. Fixed predictable costs. Zero client-side exposure.

Our Response

We Didn't Just Study The Problems.
We Fixed Them.

Every failure we documented became a feature requirement. Every vulnerability became a safeguard. Every rogue behaviour became a permission check.

We created a 28-item security backlog — categorised into what we already do better, critical fixes, quick wins, and strategic features for the roadmap.

Zack AI Security-First Design
Security Programme

Our Security Backlog

A structured approach to building the most secure AI assistant available.

7
Already Ahead
6
Critical Fixes
9
Quick Wins
12
Strategic Features

Security Audit Report 2026

We conducted a comprehensive internal security audit covering our Hosting Portal, API endpoints, authentication flows, and data handling practices. The full report is available for download.

Download Security Audit PDF
Delivered & In Progress

Security Features That Set Us Apart

Real actions taken in direct response to competitor failures.

Draft-Only Email Policy

All emails are created as drafts in the Admin UI. Mike reviews every message and manually clicks "Send" when ready. The bot can never send an email on its own. Direct sending is disabled in code.

Completed

Clear AI Identity

Zack has his own identity — zack@zackbot.ai. He never pretends to be the business owner. Every email, every message clearly comes from an AI assistant, not a human impersonator.

Completed

Human-in-the-Loop

No autonomous phone calls, social media posts, or financial transactions. Every external action goes through a human approval gate. The AI proposes, the human decides.

Completed

No Client-Side Exposure

All credentials live server-side. Nothing on the customer's machine, no config files to leak, no API keys to steal. Your laptop compromise can't touch Zack's infrastructure.

Completed

Curated Extension Library

Zero attack surface from third-party code. No open marketplace where community "skills" could hide crypto miners or backdoors. Every extension is built, tested, and maintained in-house before entering our curated library.

Completed

Fixed Predictable Costs

While competitors charge per API call (users report $250+ just for initial setup), Zack operates on a fixed monthly fee. No surprise bills, no runaway token costs.

Completed

Security Headers

HSTS, Content-Security-Policy, X-Frame-Options, X-Content-Type-Options. Hardening the web layer against injection, clickjacking, and downgrade attacks.

In Backlog

Encrypted Credential Vault

Fernet symmetric encryption for all stored OAuth tokens and API keys. Even if the database leaks, credentials remain encrypted and unusable.

In Backlog

Action Approval Workflow

Categorising every AI action as "safe" or "dangerous". Safe actions execute immediately. Dangerous actions queue for human approval. Complete transparency.

Roadmap
Under The Hood

How Draft-Only Email Works

A real example of our most impactful security feature — enforced at the code level.

zack@zackbot.ai
send-email.sh output
1#!/bin/bash
2# WARNING: Direct email sending is DISABLED
3# All emails are redirected to draft-email.sh
4
5echo "Direct sending disabled. Creating draft instead..."
6exec /root/test/email/draft-email.sh "$@"

$ ./send-email.sh -t "client@example.com" -s "Quote" -b "..."
⚠ Direct sending disabled. Creating draft instead...
✔ Draft created successfully
✔ Available in Admin UI > Emails
✔ Awaiting human review before sending

# The email NEVER leaves the server until a human clicks "Send"
# No exceptions. No overrides. No surprises.
Trust Indicators

Security Standards

Self-audited against industry-standard security frameworks and best practices.

Internal Security Audit

Comprehensive 2026 audit covering API security, auth flows, data handling, and infrastructure

HTTPS Everywhere

TLS 1.3 encryption on all endpoints. No plaintext communication, no downgrade attacks

UK Data Residency

All data stored on UK-based servers. No overseas data transfers, full GDPR compliance

Human Oversight

Every external action reviewed by a human before execution. Zero autonomous operations

Activity Logging

Every action logged with timestamps and metadata. Full audit trail for accountability

Our Journey

From Research to Reality

How competitor failures became our security features.

Discovery: Competitor Research

Deep-dived into every public complaint, CVE, forum post, and news article about Clawdbot, MoltBot, and OpenClaw. Compiled 900+ exposed instances and 386 malware-infected skills.

Analysis: Security Backlog Created

Converted findings into a prioritised 28-item security backlog. Identified 7 areas where Zack already leads, 6 critical fixes, 9 quick wins, and 12 strategic features.

Action: Security Principles Enforced

Draft-only email enforced at code level. Human-in-the-loop gates added to all external actions. Plugin marketplace deliberately excluded from architecture. Server-side credentials only.

Audit: Internal Security Review

Conducted comprehensive internal security audit. Published findings in our Security Audit Report 2026. Ongoing systematic delivery of remaining backlog items tracked in our project management system.

Your AI assistant should protect your business,
not put it at risk.

Zack AI is built with security-first architecture, human approval workflows, and zero client-side exposure. No surprises, no rogue behaviour, no runaway costs.

Book a Free 15-Minute Demo See How It Works zack@zackbot.ai 07477 133249