Confidential mandate
Education AI Safety Framework Director — Consulting
Planned Hiring / New
An education platform needs a four-month AI safety framework covering learner data, age-appropriate interaction, assessment integrity, educator oversight and incident response before scaled deployment across learner segments.
The mandate
The defined problem is that AI tutoring and assessment features are advancing without one safety standard for different learner ages, sensitive disclosures, answer integrity and educator intervention. Product-level testing does not cover ecosystem harm.
The deliverable is a learner-risk taxonomy, safety requirements, evaluation suite, red-team record, educator-oversight model, incident playbook, release gate and governance handbook.
Milestone one is due 15 October 2026 with risk taxonomy and requirements; milestone two on 30 November with completed evaluations and red-team findings; milestone three on 15 January 2027 with remediated controls, exercised incidents and accepted handbook.
The Learning Safety Council accepts when evaluations cover agreed age and subject cohorts, all critical findings are closed, educators validate intervention workflows, privacy approves data handling and two releases pass the new gate with traceable evidence.
The platform provides model configurations, system prompts, data flows, product journeys, incident history, test accounts and representative educators. Product owners remediate critical findings within agreed sprint capacity.
Why this is external work
Feature teams cannot independently red-team and approve their own releases. The client needs temporary AI evaluation, child-safety and educational-assessment expertise across disciplines. External challenge ends once client teams can operate the gate and incident process.
What you will own
- Define learner harm scenarios by age, subject, feature and level of autonomy.
- Translate risks into testable milestone-one safety requirements.
- Build evaluation sets for harmful advice, privacy, bias and assessment integrity.
- Conduct adversarial testing and classify findings by credible harm.
- Design educator oversight, escalation and learner-support workflows.
- Exercise two incident scenarios after critical remediation.
- Transfer the milestone-three release gate, handbook and evidence templates.
Candidate qualifications
- 18–22 years in education technology, AI safety, product trust or digital child safety.
- Direct governance of AI features used by children or formal learners.
- Experience building reproducible evaluations and adversarial tests.
- Knowledge of assessment integrity, pedagogy, privacy and educator workflows.
- Evidence of stopping a release because safety evidence was insufficient.
- Ability to work independently of model and platform vendors.
Non-negotiables
- No model-vendor resale or implementation commission.
- Critical safety findings close before release-gate acceptance.
- Bengaluru presence for educator validation and incident exercises.
- Test data and learner information remain in client-controlled systems.
- 49 words maximum. Which learner-facing AI release did you stop or constrain, and what safety evidence drove that decision?
- 49 words maximum. How would you build age-appropriate tests for tutoring and assessment integrity?
- 49 words maximum. Which product, model and educator inputs must be available before milestone one?
This mandate is confidential. The client is named only under a mutual NDA, and your own record is never listed, sold or shown to a company under your name until you release it for this specific mandate.