Proper Tools
Hard Problems Made Legible
Independent advisory practice on systemic infrastructure risk and long-horizon failure modes.

AI Use & Analytical Integrity

Proper Tools uses AI tools deliberately and under defined controls.

AI is not a decision-maker. It is an analytical assistant — used alongside structured reasoning, traditional research methods, and human judgment. All AI-assisted work is human-reviewed before delivery. Client data is isolated, minimized, and handled in accordance with engagement scope.

We treat AI use as a data-handling and governance question — not as a novelty, but as a natural extension of our Privacy and Security policies.

When AI Is Used

  • Synthesize research and produce structured drafts
  • Explore alternative analytical framings
  • Stress-test arguments
  • Improve clarity and communication

All AI-assisted outputs are reviewed, edited, and validated by a human before inclusion in client deliverables. AI systems are never delegated authority over conclusions or recommendations.

Model Limitations

We design around known limitations: hallucination risk, context truncation, statistical rather than causal reasoning, and bias inherited from training data.

AI output is treated as draft material, not authoritative fact. Source validation and judgment remain human responsibilities.

Client Data Handling

Cloud-Based Models

  • Dedicated accounts are created on a per-client basis.
  • Client workspaces are logically isolated.
  • No cross-client data is shared between accounts.
  • Personal or general-use accounts are not used for client-sensitive work.
  • Client-specific accounts are closed and associated working materials deleted at the conclusion of the engagement, unless otherwise agreed.

Conversation history or cached data may persist temporarily within vendor systems after account closure. We select vendors and configure accounts to minimize this residual exposure. Backend retention practices remain subject to vendor platform architecture and contractual terms.

Confidential or Regulated Data

  • Dedicated per-client locally controlled or on-premise models may be used.
  • Traditional analytical techniques are applied in parallel.
  • Sensitive datasets are not transmitted to third-party cloud AI services unless explicitly authorized by the client and contractually permitted.

Workflows are designed to minimize data exposure at all stages.

Confidentiality Controls

  • Internal lookup tables map codenames to client names, key individuals, and projects.
  • AI-assisted drafting uses these codenames rather than identifying details.
  • Identifying information is reinserted only during final human review.

This approach preserves analytical utility while minimizing disclosure risk.

Data Minimization

Only the minimum necessary information is provided to any analytical tool. If a task can be completed without sensitive identifiers, those identifiers are omitted.

Governance & Accountability

Analytical Trace Documentation

Upon request, Proper Tools can provide curated documentation of AI-assisted analytical workflows used during an engagement: representative prompt structures, redacted or codename-based excerpts, summaries of model-assisted reasoning steps, and human review annotations.

This documentation is prepared deliberately and does not constitute full raw session logs unless explicitly agreed in advance. Analytical trace documentation may be subject to additional fees depending on scope and formatting requirements.

By default, AI working materials are not retained beyond the duration necessary to complete the engagement unless otherwise agreed.

Vendor Independence

We do not submit client materials for model training. We do not opt into vendor features that use client data for training or product development without explicit contractual authorization. Where vendor terms apply, those terms are reviewed for consistency with client confidentiality obligations.

Regulatory Alignment

This policy is informed by the risk-based framework of the EU AI Act. We monitor regulatory developments and update our practices as appropriate.

AI use is treated as part of our broader security, confidentiality, and resilience posture.

Security is not the elimination of imperfection. It is the disciplined management of inevitable uncertainty. AI is governed accordingly.

Changes

If this policy changes materially, we will note it here with the date of the change.

Last updated: 2026-03-04.