Ethics statement

Our ethical commitments

KallosSim is a tool for people who work with vulnerable individuals. That places a duty on us that goes beyond compliance. This page sets out the principles that guide every decision we make about the platform.

Four guiding principles

Every product decision — what scenarios we build, how AI is used, what data we collect — is evaluated against these four principles.

Do no harm

Practice simulations involve scenarios of abuse, neglect, domestic violence, mental health crisis, and bereavement. We design the platform so practitioners can learn safely — without reproducing harm. Each practice case and persona is generated fresh by AI from human-designed parameters (roles, dimensions, rubrics, and prompt rules). We do not pre-approve every session before you see it; we rely on programmatic guardrails, user reporting, and editorial review of flagged evaluations.

Equity and anti-discrimination

A persona's name, cultural background, or protected characteristic must not determine or explain their situation. This rule is enforced in every generative prompt. AI can still slip — we ask users to report problems, sample flagged evaluations in our QA workflow, and update prompts when patterns emerge. We do not claim perfect fairness.

Digital rights

Your session data belongs to you. You have the right to access, export, and permanently delete your data at any time. We do not sell your data, share it with advertisers, or use your conversations to train AI models. Full GDPR compliance — including Articles 15 and 17 — is a baseline, not an aspiration.

Transparency and accountability

We publish what AI does in the platform, which models run, how data moves, and what prompts shape AI behaviour. If something goes wrong, we want to know. There is a named contact for ethics concerns with a five-working-day response commitment. See our AI transparency page for technical detail.

Your digital rights

Under UK GDPR and the Data Protection Act 2018, you have specific rights regarding your personal data. KallosSim is designed to make these rights easy to exercise — not buried in a support ticket.

✓
Right to access (Art. 15) You can download a full export of your session history, reflection scores, and account data at any time from your account settings. No delay, no request form.
✓
Right to erasure (Art. 17) You can permanently delete your account and all associated data from settings. Anonymisation is complete within 30 days. Aggregate anonymised statistics (with no individual identifiers) may be retained for platform improvement.
✓
No profiling for advertising Your usage data is never used to build advertising profiles or sold to third parties. The only reason we collect data is to improve your practice experience and meet our safeguarding obligations as a platform.
✓
No AI training on your conversations The conversations you have in Statutory Conversation Simulator — your responses, your reflection scores, your session transcripts — are never used to fine-tune or train AI models, whether by KallosSim or the underlying model providers.
✓
Data portability (Art. 20) All exported data is provided in a standard machine-readable format (JSON). You can take your session history to another platform, use it for CPD evidence, or share it with your employer — it's yours.
✓
Right to object (Art. 21) You may object to any processing of your data for research or platform improvement purposes at any time. Contact dpo@kallossim.com with your request.

Responsible AI

AI generates realistic practice content at scale. That power creates obligations. These are not aspirational guidelines — they are implemented rules that shape every AI call the platform makes.

Equity rules in every prompt Every AI prompt that generates a scenario, persona response, or assessment score includes a mandatory anti-stereotyping instruction. The rule requires that a persona's cultural background, name, or any protected characteristic cannot be the cause or explanation of their situation. This is enforced programmatically, not just editorially.
Human-defined structure, AI-generated sessions Practitioner roles, practice dimensions, training focuses, rubric criteria, and system prompts are designed and maintained by humans with sector expertise. Each case, persona, conversation, and score is generated by AI at runtime within those rules. There is no static scenario library in the live app and no human sign-off on every session before you practise.
Guardrails, not gatekeeping every output AI is a tool, not an authority. We constrain it with equity rules, PII filtering, evidence-first scoring, case-adjust guards, and optional post-session quality checks. Editorial staff review flagged evaluations and harm reports — not each generated case in advance. Professional judgement always takes precedence over AI feedback.
Documented prompt behaviour The prompts that shape AI behaviour in KallosSim are documented, versioned, and reviewed when models change. We do not use opaque "black box" prompting. What the AI is instructed to do is knowable and reviewable by our editorial board.
AI assessment is not a professional judgement Reflection scores and coaching feedback generated by AI are learning aids, not professional assessments. The platform labels all AI-generated feedback with a disclosure that professional judgement takes precedence. No AI output should be used as evidence of competence in formal appraisal or registration contexts.

Professional ethics alignment

KallosSim is designed to support practitioners working within established professional and statutory frameworks. Our ethical commitments are grounded in those frameworks — not separate from them.

Social work

Professional Capabilities Framework

Scenarios and dimensions are mapped to PCF domains — particularly Values & Ethics (domain 2), Diversity & Equality (domain 3), and Knowledge (domain 5). Practice simulations are designed to develop, not test, capability.

Regulation

HCPC Standards of Proficiency

The platform supports practitioners meeting Standards 1 (professional autonomy) and 10 (practise within legal and ethical boundaries). Reflection prompts are aligned to HCPC expectations for critical self-evaluation.

Statutory guidance

Working Together to Safeguard Children 2023

Children's services scenarios reflect current statutory thresholds, language, and multi-agency structures as set out in Working Together 2026. Scenario frameworks are reviewed when statutory guidance is updated.

Equality

Equality Act 2010

The nine protected characteristics under the Equality Act are explicitly referenced in our equity rules. No scenario or AI output may treat any protected characteristic as a causal factor in a service user's situation without explicit, evidence-based justification.

Alignment with these frameworks reflects current best practice. It does not constitute endorsement by the relevant regulatory bodies. Practitioners remain responsible for applying their own professional judgement in all real-world situations.

How to raise a concern

If you notice content that appears to stereotype or discriminate, output that feels inappropriate or harmful, or a practice you believe conflicts with these commitments, we want to hear from you.

Email hello@kallossim.com with a brief description of the concern. We commit to:

→ Acknowledge your email within two working days
→ Provide a substantive response within five working days
→ Review and, where appropriate, correct the content or practice raised
→ Inform you of the outcome

For data protection concerns specifically, contact our DPO at dpo@kallossim.com. For general enquiries, hello@kallossim.com.

Read our AI transparency page

Our ethics statement covers our principles and commitments. The AI transparency page covers the technical detail — which models run, what AI generates vs. what humans curate, and how we prevent bias at the prompt level.

AI transparency →

Last reviewed June 2026 · Updated when policies or architecture change