Responsible AI that earns a teacher's trust.
Cognition uses AI to help teachers see what students truly understand. Because our users include minors and our work touches academic integrity, we hold ourselves to a high standard: human oversight, fairness, transparency, and safety, in every feature we ship.
Last reviewed 8 June 2026
The principles we build to
These are not aspirations bolted on after the fact. They are the constraints we design within, and the ones we are happy for any school to hold us to.
Human in the loop
A teacher always reviews the evidence and makes the decision. The AI informs; it never judges, grades, or disciplines.
Evidence, not accusations
Cognition never returns a verdict on cheating. It surfaces what a student understands, for a person to interpret.
Fairness by design
A conversation meets each student in their own words, avoiding the style bias that detection tools impose.
Transparency
We explain what the AI does, what it cannot do, and where it can be wrong, in plain language.
Privacy and no training
Student answers and uploaded work are never used to train AI models, and we collect only the minimum needed.
Safety for minors
Conversations are guarded to stay age-appropriate and on-topic for an audience that includes children.
A conversation, not a black box
Cognition uses a large language model to do two things: generate the next question in a short, adaptive conversation, and write a clear, per-concept summary of what a student demonstrated. The model receives only the context it needs for that step, such as the task topic, the conversation so far, and, for a Defence session, the text of the student's own work. It adapts each question to the student's responses so the conversation meets them where they are.
Crucially, the model produces language, not a judgement. There is no score of guilt, no probability that work was AI-generated, and no disciplinary recommendation anywhere in the system. The teacher reads a real transcript and decides what it means.
The technology serves the relationship between a teacher and a student. The AI informs the conversation; the teacher always makes the decision.
What our AI will never do
- Decide, on its own, that a student has cheated.
- Output a probability that work was AI-generated.
- Grade, rank, or penalise a student automatically.
- Use student answers or uploaded work to train AI models.
- Profile or track students beyond the lesson in front of them.
Fairness and bias mitigation
Detection tools infer from writing style and disproportionately misclassify non-native speakers, neurodivergent students, and particular ways of writing. Cognition takes the opposite approach: it has a real conversation, in the student's own words, about the work in front of them. That design removes the central source of bias in detection. We continue to review how our questions and summaries perform across diverse student populations, and we welcome reports of unfair patterns so we can investigate and improve.
Transparency and explainability
We tell teachers and students, in plain language, that they are interacting with an AI, what it is for, and what it cannot do. Every result is grounded in an actual transcript the teacher can read in full, so the basis for a summary is visible rather than hidden inside a score. We describe our wider data practices openly in our Privacy Policy.
Accuracy and known limitations
AI systems can make mistakes. A model may misread a response, miss nuance, or summarise imperfectly. Cognition's output is evidence to inform a teacher's judgement, not a definitive measurement, and it should always be read alongside the transcript and the teacher's own knowledge of the student. We are clear about this limitation precisely because the stakes in education are real: the output must never be the sole basis for a consequential decision about a student.
Safety guardrails for minors
Because many users are children, conversations are guided and filtered to stay age-appropriate, respectful, and on-topic. We apply content moderation, keep the conversation focused on the educational task, and design prompts to be supportive rather than adversarial. A Cognition conversation should feel like an encouraging discussion, not an interrogation.
Privacy and no training on student work
We collect the minimum information needed to run a conversation, and students never create accounts or provide emails or passwords. Student answers and uploaded work are never used to train or fine-tune AI models. We configure our AI processing so that content is used only to deliver the service, and we require providers to meet that standard. Full detail is in our Privacy Policy.
Model providers and oversight
Cognition is built so the underlying AI provider can change over time as the technology improves. Whichever provider we use, we require the same commitments: no training on your content, processing limited to delivering the service, and appropriate security and data handling. We select and review providers with these requirements in mind.
Our philosophy on academic integrity
Cognition reframes integrity from policing to understanding. Rather than trying to catch students, we give them a chance to demonstrate what they know, and we give teachers genuine evidence of learning. Asking a student to explain, defend, and extend their reasoning also happens to be one of the most effective ways to deepen learning, so an integrity check becomes a moment of teaching.
Alignment with recognised frameworks
Our approach is informed by widely recognised principles for trustworthy AI, including human oversight, fairness, transparency, accountability, privacy, and safety, as reflected in frameworks such as the OECD AI Principles and UNESCO guidance on AI in education. We treat education as a high-responsibility setting and design accordingly.
Accountability and governance
Responsibility for our AI practices rests with Perspective Global Pty Ltd. We review our prompts, safeguards, and providers as the product and the technology evolve, and we update this page when our practices change. If a feature cannot meet our standards for fairness, safety, or transparency, we do not ship it.
Reporting a concern
If you encounter output that seems unfair, unsafe, inaccurate, or inappropriate, please tell us so we can investigate and improve. Educators, students, and parents can all raise concerns, and we treat reports about safety and fairness as a priority. Email hello.cognition@pixelverse.tech, or privacy.cognition@pixelverse.tech for privacy-specific questions.
Reviewing Cognition for your school?
We are glad to walk through how our AI, privacy, and safety practices fit your requirements.