Find anything
AI Use Policy
Certifiedge supports responsible AI use when it helps people learn, plan, practise or improve their work. Material AI assistance should be disclosed where it affects the evidence a reviewer is being asked to trust, and private or confidential information should be protected. Learners remain responsible for checking accuracy, explaining their decisions and using human judgment where the task or decision requires it.
AI use philosophy
AI can help learners draft, organize, summarize, plan and improve work when the task allows it. The value comes from how well the learner questions, checks, edits and applies the output rather than from polished wording alone. Certifiedge therefore looks for visible human judgment, appropriate disclosure, careful verification and safe handling of information.
Human judgment requirement
Learners remain accountable for the work they submit, even when an AI tool helped produce part of it. A reviewer may ask the learner to explain choices, correct weak claims or show how an output was checked. Important decisions should rely on evidence and accountable human review rather than treating AI output as the final authority.
Source-grounded AI support
When Certifiedge AI uses approved learning material, the experience can show the source titles that informed the response so you can check the original lesson or resource. Source access follows the same learning access rules as the rest of Certifiedge, so private or paid learning material should not appear simply because an AI request asks for it. If useful sources are not available, the assistant should say so or provide general study guidance instead of pretending to have evidence it does not have.
AI support should only use learning sources your account is allowed to access. Organization learning stays within the appropriate organization context, and personal learning access remains tied to the learner who holds it. This helps keep useful personalization from becoming a shortcut around privacy or access controls.
A helpful answer should make it easier to return to the lesson, policy or approved resource behind an important claim. Source labels are a starting point for verification, not a substitute for reading the full material when the detail matters. Assessment decisions, credentials and other consequential outcomes continue to rely on their own governed evidence and human-review processes.
Passwords, access tokens and other account secrets should never be entered into AI support. Highly sensitive personal information is kept away from external model services through Certifiedge safeguards, and live-model use can be limited further by organization policy. When a live service is unavailable or not appropriate for the information involved, Certifiedge can fall back to local guidance rather than lowering the protection boundary.
One front door, separate authorities
Ask Certifiedge can identify whether a request is best handled as learning support, source-grounded explanation, planning, career exploration, account guidance or a supported action proposal. The route is selected before model output is trusted to make authorization choices, and only learning sources the signed-in account is permitted to use can become grounded context. This keeps one simple conversation experience on the surface while preserving separate controls underneath.
On each turn, eligible users can decide whether recent conversation context and explicitly saved memory may be used. A request to send or create something can only prepare a Connected Action; it cannot bypass the exact-action approval screen. Ask Certifiedge also cannot accept evidence, award mastery, finalize assessment, declare readiness, issue credentials, decide admission or change learning access.
How agentic AI gets its facts
Certifiedge AI can coordinate specialized learning, assessment-interpretation, evidence, credential, organization and support roles. Before a model answers, the system retrieves only the authorized, minimized facts needed from the Certifiedge services that already own those records. Canonical operational records outrank published knowledge, confirmed AI memory and model prior when they conflict.
Raw assessment responses, private evidence artifacts, passwords, safeguarding records and financial secrets are not ordinary AI context. The assistance record keeps hashes and authority metadata so the system can show what shaped an answer without turning the model into a second database. Whether the reasoning provider is local or OpenAI, it cannot silently accept evidence, award mastery, determine readiness, issue credentials, make employment decisions or change financial authority.
Memory and continuity
Certifiedge separates short-term session continuity from longer-term memory. Recent session context can be retained for a limited period so you do not have to repeat the same background, while longer-term memory remains a separate adult opt-in and is saved only after explicit confirmation. You can review, remove or disable what is remembered, and highly sensitive personal information or account secrets should never be placed in longer-term AI memory.
Remembered context can make assistance more relevant, but it never becomes assessment truth, evidence, mastery, readiness, a credential or permission to act on your behalf. For child accounts, longer-term memory stays off. Actions in other apps or services require a separate, clear permission step; conversation or memory alone is never enough.
Actions require a separate decision
Certifiedge may prepare a narrowly supported action for an eligible signed-in user, but a suggestion, conversation, saved memory or account-level action toggle never authorizes the action itself. Before anything with a real side effect happens, Certifiedge shows the exact action details and asks that user to approve or reject that specific request. Actions with financial, credential, account-privilege, destructive-record, browser, shell, file-system or arbitrary web-access authority are not available through this release.
When an approved action succeeds, Certifiedge records a receipt so the user can see what happened. A failed or expired request is not silently retried, and changing the requested details requires a new approval. Learning and assessment decisions remain with the systems and accountable people that already own those decisions.
FAQ
No. AI permissions depend on task rules, confidentiality risks and assessment requirements.
Yes, when disclosure is complete, quality checks are clear and human judgment is visible.
The submission may require revision or rejection depending on risk and quality impact.
No. Responsible, permitted and documented AI use can support learning and work quality. Undisclosed, unsafe or prohibited use may require revision or further review before a learner progresses.
Where AI may be allowed
Where AI is restricted
Disclosure rules
Privacy and confidential data
Before using an AI tool, learners and teams should check whether the material contains personal, confidential, commercial or regulated information. Sensitive information should be removed or protected unless an approved secure method clearly permits that use. When in doubt, use a safer example or ask the responsible organization before sending the information to an external service.
Responsible AI use includes protecting confidential information, being truthful about material assistance and checking important claims against reliable sources or direct evidence. The learner should be able to explain what changed after the AI output and why the final version was accepted. This keeps AI support visible without confusing assisted production with unaided competence.
AI assistance does not make fabricated references, misleading citations or unattributed copying acceptable. Work that hides material assistance or presents invented information as fact can be returned for correction or rejected under the applicable assessment rules. The safest approach is to verify sources, disclose significant assistance when required and make sure the final submission reflects the learner’s own accountable judgment.
Use AI only where the task permits it and disclose material assistance honestly when required. Check the output, document meaningful changes and protect confidential information throughout the process. Be ready to explain the final work in your own words and defend the decisions you made.
Review the evidence for quality, relevance and practical judgment rather than rewarding polished wording by itself. Check whether material AI assistance was disclosed and whether the learner can explain, verify and improve the output. Where the task requires unaided performance, keep that boundary explicit and assess the learner on the evidence produced under those conditions.
Employer and partner interpretation
AI-assisted evidence can still be useful when the process is clear and the learner’s contribution can be understood. Employers and partners should review the work, the learner’s judgment, relevant quality checks and any required disclosure before relying on the result. For consequential decisions, this evidence should sit alongside role-specific review and accountable human judgment.
Consequences of misuse
Consequences should remain proportionate, documented and reviewable.
Policy note
Certifiedge Internally Approved — Deployment Activation Pending