AI-enabled phishing and scams
Generated emails, SMS, social messages, fake support chats, and urgent requests can look polished and personal. Build habits that verify requests through trusted channels.
AI Security Hub • defensive guidance
Practical protection for the AI threat era
A focused hub for defending people and organisations against AI-enabled scams, deepfakes, voice fraud, prompt injection risk, unsafe AI tool use, data exposure, and incident pressure.
AI Scams
Deepfakes
Prompt Injection Awareness
Safe AI Use
AI Governance
Core habitVerify identity before trust
Data ruleClassify before you paste or upload
Response rulePreserve evidence, contain, recover
AI threat lens
What the hub helps you defend against
AI changes how believable, fast, and personalised attacks can become. The response is not panic; it is better verification, safer data handling, least privilege, monitoring, and response discipline.
Deepfake and voice fraud
Audio and video impersonation can pressure people into payment changes, sensitive disclosures, or access approvals. Verify unusual requests outside the original channel.
Prompt injection awareness
AI tools that read web pages, documents, email, or tickets can be influenced by hidden or malicious instructions. Treat AI output as untrusted until reviewed.
Sensitive information disclosure
Prompts, file uploads, chat history, plugins, and connectors can expose confidential information if people paste or connect data without clear approval.
AI agent permissions
Agents that can browse, email, create files, call APIs, or make changes need least privilege, approval steps, logging, and a clear human decision point.
Governance and incident readiness
Organisations need ownership for AI acceptable use, data classification, vendor review, escalation, and evidence handling when something goes wrong.
AI safety checklist
Start with simple rules that reduce real risk
Safer AI adoption starts with clear boundaries. People should know what they can use AI for, what information must never be pasted, when to verify outputs, and who to contact if a tool behaves unexpectedly.
This checklist is written for practical use by individuals, small businesses, leaders, and technical teams. It is defensive guidance, not legal advice or a replacement for formal security assessment.
Safe AI use
A practical control model for AI adoption
HD Cyber Defence recommends a calm, control-based model: know where AI is used, classify the data, limit permissions, verify outputs, monitor risky signals, and prepare a response path.
1. Discover
Identify AI tools, browser extensions, copilots, chatbots, automations, connectors, and local agents already in use.
2. Classify
Separate public information from confidential, personal, financial, health, security, source code, and board-level material.
3. Limit
Apply least privilege to connected tools, accounts, folders, mailboxes, APIs, and workflow automation.
4. Verify
Review AI output before decisions, payments, access approvals, customer responses, code changes, or security actions.
5. Monitor
Watch for unusual access, unexpected sharing, suspicious browser extensions, leaked credentials, and impersonation attempts.
6. Respond
Prepare escalation, evidence capture, containment, communication, and recovery steps for AI-related cyber incidents.
Role-based guidance
Different audiences need different AI defence messages
The same threat may need a household explanation, a small-business checklist, an executive risk briefing, or a technical control review. The hub gives each audience a safe starting point.
Protect accounts, verify family emergency messages, recognise voice-clone pressure, and secure recovery options.
Verify supplier bank changes, protect shared mailboxes, manage AI tools, and prepare basic incident response steps.
Ask who owns AI risk, what data is exposed, how AI use is approved, and how incidents will be escalated.
Review integrations, logs, permissions, prompt injection exposure, data flows, and monitoring coverage.
Pause before urgent requests, verify identity, avoid sensitive prompts, and report suspicious messages early.
Validate outputs, isolate tools, protect system prompts, test retrieval sources, and document human approval points.
Recommended first steps
Build AI defence from everyday habits
Strong AI security is not only a technology purchase. It is a repeatable operating habit that combines policy, people, process, monitoring, and response.
Write the rule
Define what staff can use AI for and what information must stay out of public tools.
Verify the request
Use trusted-channel confirmation for payments, access changes, executive instructions, and sensitive disclosures.
Limit the agent
Keep AI agents and connectors on minimum permissions with review points before action.
Train the people
Teach realistic AI phishing, voice fraud, deepfake pressure, and safe prompt behaviours.
Watch the signals
Monitor suspicious logins, unusual sharing, leaked credentials, exposed services, and risky tool adoption.
Prepare the response
Have a playbook for suspected account compromise, data exposure, impersonation, and AI tool misuse.
Use the hub
Make AI defence practical, not overwhelming
Use this hub as the starting page for AI cyber safety conversations, staff awareness, executive briefings, and safer AI adoption planning.