Personal AI Safety Module

Core Principle

Safety is a continuous responsibility

Installing a safety layer is not the end of the process. The module guides you through an ongoing cycle — not a one-time checklist.

Step 1

Before you install

The module begins with guided preparation — prompting you to understand your own AI environment before making any changes. The assessment adapts to your circumstances.

• What do you use AI for?
• What information can your AI access?
• What tools, services, accounts, files, or devices are connected?
• What actions can the AI take?
• What consequences could those actions create?
• What are your biggest concerns?
• What risks are already known?
• What risks or blind spots may not yet be visible?

Step 2

Install & configure

The installation experience guides rather than simply downloads. The module accommodates a wide range of environments — from personal AI assistants and chatbots to agents, local models, cloud systems, and multi-agent environments.

No two configurations are the same. The module adapts to your system, your connections, and your risk profile.

• Select an appropriate user or system profile
• Identify the AI system and environment
• Map relevant connections and external dependencies
• Establish permissions and boundaries
• Identify actions requiring confirmation
• Establish user preferences
• Review the initial configuration
• Confirm before activating consequential controls

Step 3

Understand your AI system

The module establishes an initial safety baseline — examining your AI environment across a range of risk categories. It explicitly searches for blind spots rather than only checking predefined risks.

Central Feature

Find what you may not have seen

When the module identifies a possible blind spot, it does not simply display “Risk detected.” It presents a structured resolution flow — keeping you informed and in control at every step.

1. 01
   BLIND SPOT

   A possible concern is identified.

2. 02
   WHAT WAS OBSERVED

   Plain-language description of the finding.

3. 03
   WHY IT MATTERS

   Potential significance to the user.

4. 04
   POSSIBLE CONSEQUENCES

   What could happen if unaddressed.

5. 05
   EVIDENCE / BASIS

   The information supporting the finding.

6. 06
   UNCERTAINTY

   What is not yet known or confirmed.

7. 07
   POSSIBLE RESOLUTIONS

   Several paths the user may consider.

8. 08
   RECOMMENDED NEXT STEP

   A suggested starting point.

9. 09
   USER DECISION

   You decide what to do.

10. 10
    IMPLEMENTATION

    The chosen resolution is applied.

11. 11
    VERIFICATION

    Confirm the resolution worked.

12. 12
    REASSESSMENT

    Update the safety picture.

Assessment Language

Beyond safe or unsafe

The module does not force every observation into a simple binary. It uses appropriate language depending on the actual assessment — including the ability to state that it does not have enough information to reach a conclusion.

Ongoing Operation

Safety does not end after installation

The module is designed to support continual reassessment as your AI systems, connections, and circumstances evolve. The scope of monitoring depends on the environment and what the implementation actually supports.

• Monitoring relevant activity
• Detecting meaningful changes and new external connections
• Detecting permission changes or unusual behavior
• Reassessing when the AI system or your circumstances change
• Reassessing after incidents or significant updates
• Providing new safety guidance as the system evolves

Design Principle

Human authority remains central

The safety layer is intended to support human understanding and control. It should not silently make consequential decisions on behalf of the user.

Where a consequential action requires human judgment, the system explains the situation, presents relevant options, and requests appropriate confirmation. The precise level of autonomy depends on your configuration and the risk associated with the action.

Foundation

Built on established safety practice

Safe & Sane’s underlying philosophy is derived from its broader safety engineering work and incorporates relevant established industry good practices. The objective is to combine established practice with the evolving Safe & Sane framework — not to replace existing professional disciplines.

Open by Design

Open by design

The Personal AI Safety Module is intended to remain open source. People should be able to inspect the architecture, examine the source, test it, identify weaknesses, propose improvements, and contribute where appropriate.

Openness itself has responsible boundaries where security, privacy, responsible disclosure, or other legitimate constraints require them.