Ethics and Supervision Policy for Artificial Intelligence

1. Purpose and scope

This policy governs the responsible and ethical use of computer technologies and artificial intelligence in the context of:

• documentary research and the organization of evidentiary data;
• the drafting of citizen submissions, briefs, press releases and open letters;
• electronic communications and public correspondence;
• tools and services deployed on the williamguindon.me site.

2. Terms and framework for using AI infrastructures

I provide strict and functional supervision of IT technologies. Artificial intelligence tools are used exclusively as technical support instruments, documentary research and software assistance, in order to compensate for the substantial asymmetry of resources in the analysis of environmental files.

2.1 Documentary analysis and exploration of complex corpora

I use indexing, semantic search and data processing technologies (notably full-text search engines and cutting-edge language models) to identify, categorize and cross-reference evidentiary passages within voluminous documentary sets (hydrogeological studies, authorization orders, technical reports). These tools only intervene in the localization of textual elements; the legal qualification, the interpretation of the facts and the writing of theses remain under my exclusive human appreciation.

2.2 Running local models for common tasks

For elementary linguistic operations that do not require any remote computing power — such as syntactic revision, optimization of sentence structures and occasional textual corrections — I give priority to the Bonsai 27B model (derived from the Qwen open source architecture), executed entirely locally on my computer hardware. This protocol ensures the strict confidentiality of the data processed and prevents any unnecessary transfer of requests to third-party servers.

2.3 Technical assistance for source code and web tools

I use models to assist in the development, debugging and standardization of programming files (HTML, style sheets, functional scripts) hosted on my domain. Each generated software or algorithmic component is subject to a functional audit and manual validation prior to its integration into production.

2.4 Proportionality and environmental assessment

The energy consumption induced by the hosted calculations (estimated at a few tens of kilowatt hours on the Hydro-Québec network, which is largely carbon-free) meets a criterion of material proportionality. This recourse is limited to steps aimed at documenting and preventing permanent and large-scale environmental damage (in particular the protection of wetlands and the control of dangerous materials).

3. Prohibited uses

1. Publish a text, image, sound or video substantially generated by generative AI without identifying them as such (principle 14.7 of the Quebec Press Council).
2. Present unverified AI output as verified fact (models can generate factual errors).
3. Using AI to fabricate evidence, alter documents or create false testimonies.
4. Transmit personal, confidential or sensitive information (sources, internal documents, data of minors) to an AI model.
5. Relying on AI for editorial decisions or to distort source statements.

4. Human editorial control — procedure

Before any publication or sending:

• The final content is written or edited by William from verified sources;
• Any AI output used is reread, cross-referenced with the sources and corrected by William;
• Facts, figures, quotes and documents are checked against the original sources, never against AI output alone;
• Social media posts and emails prepared with AI assistance are always double-checked by William before sending/publishing;
• Any published content substantially generated by AI is visibly and inseparably identified.

5. Transparency, attribution and primacy of human inquiry

Automated systems are by nature incapable of carrying out a legal investigation process, determining an evidentiary strategy or understanding the material reality of the premises. The entire investigation of my files is based on my personal investigation work:

• Archive examination and evidentiary conservation: I myself carry out the identification in the public archives of historical documents, judgments or court reports not indexed by the usual search engines, of which I ensure the manual capture, temporal authentication, classification and inclusion in the proof annexes.
• Exercise of rights of access to information: I personally ensure the formulation, legal qualification, formal filing and monitoring of requests for access to documents from public bodies as well as related administrative appeals with provincial and federal authorities.
• Direct obtaining of technical documents: I obtained and assembled administrative correspondence, specialized hydrogeological studies and engineering plans not publicly disseminated by direct approaches to information custodians.
• Physical findings in the field: I conducted the in situ examination of topography, receiving watercourses and threatened wetlands, while personally cross-checking these observations with geospatial data and consulting with independent scientific experts.
• Counter-expertise and detection of anomalies: I myself identify methodological deviations, toxicological anomalies and excesses of standards in industrial and government reports.
• Legal theory and formulation of appeals: I developed all of the legal argumentation in support of my efforts (in particular under the mechanisms of ACEUM (Canada–United States–Mexico Agreement)/CUSMA). AI does not intervene at any time in the design of the theses defended.
• Methodological transparency: My use of data processing technologies is fully explained. It constitutes a method of technical organization proportionate to ensuring the exhaustive processing of thousands of regulatory pages.
• Traceability and opposability by primary sources: The validity of my work is based on the probative force of officially referenced official acts, engineering studies and legal provisions. No assertion arises from autonomous algorithmic processing not corroborated by a verifiable material source.
• Accountability and personal editorial responsibility: Any content made public or filed before a court or an international organization engages my exclusive personal responsibility, without intellectual delegation or release of liability.

6. Liability

The site remains solely editorially responsible for all published content, including if an error comes from an AI tool. The use of an AI tool is not an excuse and does not waive the rules of truthfulness and accuracy.

7. AI Documentary Assistant — Formal exclusion of liability, independence and absence of legal value

Any person accessing or conversing with the AI ​​Documentary Assistant expressly acknowledges and agrees to the following provisions:

• Absolute statutory distinction: The AI ​​Documentary Assistant is an automated algorithmic system and has no status as agent, spokesperson or representative of William Guindon, his relatives or his technical advisors. Statements or responses produced by the assistant in no way reflect the opinions, wishes or statements of William Guindon.
• Total absence of legal advice or evidentiary authority: This automated assistant does not replace professional legal advice, legal advice from an attorney or official scientific analysis. The summaries generated have no legal or probative value before the Commission de coopération environnementale (CCE / ACEUM), the courts, the BAPE (Québec Bureau of Public Environmental Hearings) or any government ministry or agency. Only briefs, decisions and supporting documents officially listed and filed in the register are authentic.
• Strict confidentiality & Prohibition of posting sensitive data: Do not submit any confidential information, personally identifiable information, identifying information, sensitive documents or trade secrets in this chat. Conversations are processed by software interfaces (local browser or language model API). William Guindon cannot be held responsible for the information that users voluntarily choose to enter in the chat interface.
• Full exclusion of liability: William Guindon, his relatives and his collaborators disclaim any direct, indirect or subsidiary liability for any factual errors, omissions, approximations, algorithmic hallucinations or erroneous conclusions generated by the AI ​​Documentary Assistant, as well as for any decision or action taken on the basis of its responses.

8. Absolute confidentiality and non-consultation of personal API keys (BYO-API)

• Storage exclusively confined to the user's terminal: Your API keys and local instance addresses (Ollama, LM Studio) are saved exclusively within your own browser's local sandbox (multi-level localStorage, IndexedDB and CacheStorage to resist session cookie dumps without ever leaving your machine).
• Direct and encrypted client-provider connection: The requests made pass directly and without intermediary from your browser to the official gateways of AI providers (e.g. generativelanguage.googleapis.com, api.openai.com, api.anthropic.com, api.mistral.ai, openrouter.ai). No proxy, mirror server or log managed by William Guindon intercepts, sees or logs your identifiers or the content of your exchanges.
• Contrôle et droit à l'oubli immédiat : Vous pouvez à tout instant visualiser votre clé en clair grâce au bouton d'affichage (), la modifier ou la purger irrévocablement de votre navigateur via le bouton « Oublier ».

This policy is a stand-alone policy of williamguindon.me, published for public transparency. No signature is required.