AI Transparency Statement

Our Use of Artificial Intelligence

MHLE uses artificial intelligence to help users analyze educational content, organize knowledge, generate study materials, explore alternative perspectives, transcribe audio, produce spoken content, retrieve related material, and verify selected factual claims.

This statement explains where AI is involved, what information may be processed, the limits of AI output, and the safeguards MHLE applies. The precise feature set depends on account type, role, age, organization configuration, and feature flags.

AI Capabilities

MHLE may use AI for functions such as:

• Multi-perspective or "multi-headed" analysis of notes and uploaded material.
• Summaries, study aids, assessments, simulations, and learning artifacts.
• Socratic coaching and retrieval of relevant source paragraphs.
• Semantic embeddings, similarity search, gap analysis, and knowledge relationships.
• Audio transcription and text-to-speech.
• Standards extraction, differentiated educational content, and pacing support.
• Moderation, classification, persona-aware presentation, and safety checks.
• Research or factual verification where that feature is enabled.
• Administrative, marketing, reporting, and customer-success assistance where authorized.

Not every result shown in MHLE is produced by AI. Some scores, limits, filters, workflow states, and reports are calculated through deterministic application rules.

Providers and Model Selection

MHLE integrates with third-party providers that may include OpenAI, Anthropic, Google, and Perplexity. Different providers may be selected for different operations based on task fit, availability, safety, quality, latency, configuration, and cost.

The current provider and processing inventory is maintained in the Subprocessor List. Model names and versions may change as providers retire, rename, or update services. Material provider changes are reviewed through MHLE's vendor and AI governance processes.

Information Sent for AI Processing

Depending on the requested feature, information sent to a provider may include:

• Note text, titles, document excerpts, course or standards content.
• A user's instructions, question, selected learning profile, or contextual preferences.
• Instructor-entered observations or other authorized institutional content.
• Audio supplied for transcription.
• Text supplied for speech synthesis.
• Text fragments used to produce embeddings or semantic matches.
• Factual claims or topics supplied for verification.

MHLE aims to limit provider requests to information needed for the operation. Users should not submit unnecessary sensitive information, credentials, government identifiers, medical records, or other data they are not authorized to process.

Training and Provider Use

MHLE's current production policy is not to use customer or student content to train its own general-purpose AI models. MHLE uses provider business/API services for requested processing and seeks configurations and contractual terms that restrict provider training use.

Processing by an AI model is not the same as training that model. The Data & Training Policy explains this distinction and the current policy in more detail.

Human Role and High-Impact Decisions

AI output is assistive. It may support learning, reflection, planning, or authorized administration, but it should not replace human judgment.

MHLE does not intend AI output to be the sole basis for:

• A final grade, admission, discipline, expulsion, or academic-integrity determination.
• Employment, credit, housing, insurance, or legal eligibility.
• A medical, mental-health, financial, or legal decision.
• A determination of disability, special-education eligibility, or another protected status.

Educators and authorized decision makers are responsible for reviewing relevant source material, context, and policy before acting on AI-assisted information.

Accuracy and Known Limitations

AI systems can:

• Generate statements that sound plausible but are incorrect.
• Omit important context or misunderstand ambiguous material.
• Reflect bias present in data, prompts, evaluation sets, or human feedback.
• Perform unevenly across languages, reading levels, cultures, and academic disciplines.
• Produce different answers to similar requests.
• Rely on provider knowledge that is incomplete or out of date.

Users should verify citations, calculations, quotations, legal or policy claims, and other important facts against authoritative sources.

Safeguards and Evaluation

Depending on the feature, MHLE applies safeguards that may include:

• Input safety checks and content moderation.
• Age- and persona-aware prompting or output constraints.
• Feature flags and controlled rollout.
• Output sanitization before browser rendering.
• Usage limits and cost attribution.
• Error handling and provider fallback logic.
• User feedback and content-reporting paths.
• Regression, adversarial, safety, and quality evaluation.
• Model inventory and periodic benchmark review.
• Logging designed to support troubleshooting and accountability while limiting exposure of sensitive content.

See the AI Quality & Safety Testing Summary. Published targets are program objectives; they should not be interpreted as a warranty that every output meets the target.

Children's and Student Data

AI processing of institution-managed student data is limited to authorized educational purposes and the institution's instructions. MHLE's published policy prohibits using student data for advertising or general-purpose model training.

Accounts identified as belonging to a child under 13 are subject to parental, guardian, or school authorization controls before content-creation and AI features become available.

User Controls and Questions

Users may choose whether to invoke many AI-enabled features, review generated content before using it, report problematic content, and use applicable privacy or deletion controls. Institutional administrators may have additional configuration and contractual options.

• AI or technical questions: contact details withheld pending approval
• Privacy questions: contact details withheld pending approval
• Institutional assurance requests: contact details withheld pending approval

Review Note

This statement remains Draft until the current production model inventory, provider data terms, high-impact-use restrictions, and contact addresses are formally approved.