AI Disclosure

ChefSphere uses artificial intelligence ("AI") in several features. This document explains which features are AI-powered, what kinds of AI models we use, how your inputs and outputs are handled, and the limits you should understand before relying on AI output. It supplements the Terms of Service and the Privacy Policy. The disclosure below is designed to meet Apple App Store Review Guideline 5.1.2 and the transparency obligations of Article 50 of Regulation (EU) 2024/1689 (the EU AI Act), which require that: (i) users are informed they are interacting with an AI system, (ii) AI-generated or AI-modified content that could reasonably be mistaken for authentic human content is labelled, and (iii) operators do not deploy AI in the ways prohibited by Article 5 of the AI Act.

AI Act risk classification. None of the AI features described below fall within the list of prohibited practices in Article 5 of the AI Act, and none are high-risk AI systems within the meaning of Article 6 and Annex III of the AI Act — they are not used for biometric identification, critical infrastructure, education scoring, employment, access to essential services, law enforcement, migration/border, or the administration of justice. They are general-purpose transparency-obligation systems under Article 50.

1. Which features use AI

• AI Chef (conversational assistant) — a chat assistant that answers cooking, nutrition, meal-planning, budget, and technique questions.
• Food and fridge photo analysis — image-understanding that identifies ingredients, dishes, or shopping lists from photos you submit.
• Preference and dietary matching — algorithms that rank recipes and meal plans against your stated preferences, dietary rules, and goals.
• Budget and grocery-price optimisation — suggestions that factor in the prices you and the community have shared.
• External recipe URL extraction — extraction of structured recipes from third-party web pages.
• Personalised meal-plan generation — generation of weekly and monthly meal plans matched to your preferences and available ingredients.
• Automated content moderation assistance — classifiers that flag posts, images, comments, or listings that may violate the Acceptable Use Policy. Classifier signals are reviewed by humans before account-level actions.
• Official AI Characters — ChefSphere-operated AI personas that may publish cooking inspiration, recipes, grocery-price content, Chef Lives slots, or community posts. The active public roster is listed at /about/ai-characters.

2. What AI is and is not

Every feature above is built on statistical pattern prediction, not on understanding. AI output can look fluent and confident while being wrong. In particular:

• AI output is informational only. It is not medical, dietary, financial, legal, or nutritional advice. Do not rely on it for decisions that affect your health or safety. If you have allergies, intolerances, a medical condition, or you are pregnant, always verify information with a qualified professional.
• AI can be confidently incorrect ("hallucinate"). Treat cooking times, temperatures, allergen flags, and nutritional claims in AI output as starting points to verify, not final answers.
• AI can reflect biases present in its training data. We take active steps (see Section 7) to minimise this, but we cannot fully eliminate it.

3. Providers and model categories

We work with several third-party AI / machine-learning providers (category-level disclosure; GDPR Art. 13 allows category-level naming of processors). The categories currently in use are:

• Large language models (text reasoning) for AI Chef and preference matching.
• Vision / multimodal models for fridge and food photo analysis.
• Structured-extraction models for the recipe URL extractor.
• Moderation classifiers for text and image safety.

Every provider is engaged as a processor under a written agreement that: requires confidentiality and security safeguards compatible with the Privacy Policy; prohibits the use of your inputs and outputs to train the provider's general models unless you have given explicit consent to that specific processing (we currently do not enable such training); restricts retention to what is necessary for the stated purpose; and supports data subject rights.

If you want to know the specific vendors in use on the date of your request, write to [email protected]. We publish the list by category because vendor names can change without any change to the personal data processed.

4. Data sent to AI providers, retention, and training

• What is sent. For each AI interaction we send the minimum input needed to produce an answer: the prompt or image you gave us, relevant profile context only where you enabled it (for example dietary preferences for recipe matching), and short-term conversation context for multi-turn chats.
• Personal identifiers. We do not send your name, email, payment data, or precise location to AI providers.
• Short-term retention. Providers may keep inputs and outputs for a short operational window (typically up to 30 days) to investigate abuse, safety, or billing. After that, they are deleted or anonymised.
• No model training. We do not permit AI providers to use your inputs or outputs to train or fine-tune their general models. If that ever changes, we will disclose it in this document and, for any processing that requires it, ask for your explicit consent before it begins.
• Your local history. Chat and image history is stored in your ChefSphere account and is visible only to you unless you explicitly share a reply. You can delete your AI history at any time from in-app settings.
• Explicit third-party AI disclosure and consent — Apple Review Guideline 5.1.2(i) (November 2025). This Disclosure, read together with the first-use consent prompt shown in-app on every AI feature, satisfies Apple's requirement that personal data shared with third-party AI providers is clearly disclosed and that explicit permission is obtained before sharing. You can withdraw that permission at any time by opening the Chef AI → Settings page in the app, which exposes the full list of granular data-sharing toggles (food preferences, cooking history, nutritional data, health data, and more) and lets you disable any AI feature without affecting the rest of the Service. If a future update introduces a new AI provider or a new category of data sent to an existing provider, a fresh consent prompt is shown before the first use of the changed feature.

5. Quotas and sustainable use

AI features have usage quotas that vary by plan tier (Free / Premium / Pro) and by feature. Quotas are there to keep the Service fast and affordable for everyone, not to limit access unfairly. Your current quotas and usage are visible in the app. Running out of a quota does not break core features; it only pauses further AI use until the quota resets or you upgrade.

6. Safety and consent

Several safety controls apply automatically:

• Age-appropriate use. AI Chef does not respond to questions that are inappropriate for users below the platform minimum age (see the Children & Age Policy).
• Harm-reduction filters. AI Chef is configured to decline or redirect requests that may cause physical harm (for example eating disorders, unsafe food handling, weaponisation) and to guide the user to proper resources.
• Allergy and intolerance awareness. If you have declared an allergy or intolerance in your profile, AI features prefer recipes and suggestions that avoid the flagged ingredient, but you must always verify the final result.
• Sensitive topics. AI Chef avoids medical, dietary, or mental-health diagnoses and will recommend contacting a qualified professional.

For features that process special-category data (health, nutrition), we ask for your explicit consent at first use. You can withdraw consent at any time from in-app settings; past processing remains lawful until withdrawal.

7. Bias, quality, and human oversight

We monitor AI output for systematic errors and bias through internal evaluation sets and user reports. You can report a problematic AI response from the chat interface; a human reviews reports that affect moderation, safety, or account standing. For any automated moderation action that restricts your content or your account, you can request human review by writing to [email protected] with your reference ID.

We do not use AI to make solely automated decisions that produce legal or similarly significant effects on you within the meaning of Art. 22 GDPR.

8. AI-generated and AI-modified content you publish

Under Article 50(2) of the EU AI Act, providers of AI systems that generate synthetic audio, image, video, or text must ensure the outputs are marked as artificially generated or manipulated in a machine-readable format where technically feasible. Under Article 50(4), deployers of an AI system that generates or manipulates content constituting a deepfake must disclose that the content has been artificially generated or manipulated.

In line with those obligations:

• Synthetic content we generate. Recipes, meal plans, and reply text produced by AI Chef are stored in your account with a machine-readable marker that identifies them as AI-generated. If you copy that text into a community-visible area, our user interface shows an "AI-assisted" label and a tooltip pointing to this disclosure.
• Official AI Character content. Content published by an official ChefSphere AI Character is attributed to that character, carries a persistent visible disclosure line, and is stored with AI attribution fields. ChefSphere does not hide AI Character content behind fake human accounts.
• Synthetic content you publish. If you publish AI-generated or AI-modified content in a community-visible area (a post, a community recipe, a profile image, an ebook sample), you must declare the substantial use of AI — either by using the "AI-assisted" toggle in the composer or by stating it in text. We may add the label automatically when we have reliable signals that the content is AI-generated. Undeclared AI content that could reasonably be mistaken for authentic human content violates the Acceptable Use Policy.
• Deepfakes are prohibited. You may not upload or share AI-generated or AI-manipulated images, audio, or video that depict a real, identifiable person in a way that could mislead a reasonable viewer about what that person said, did, or endorsed. This prohibition is independent of the labelling duty above and applies even if the content is clearly labelled as AI.
• Art and satire exception. The deepfake disclosure obligation does not apply to content that is evidently creative, artistic, fictional, analogous, or satirical, provided it does not impair the rights of identified persons. The prohibition on defamatory or misleading deepfakes continues to apply.

9. Open research and red-teaming

We maintain an internal responsible-AI checklist that is reviewed before material changes to AI features. External researchers who find genuine safety issues can report them responsibly to [email protected].

Official AI Character launches are reviewed through an internal approval workflow. Generated assets must pass moderation before an awaiting-review content plan can be approved, and publish logs record which admin approved and published each item.

10. Your rights

You have every right described in the Privacy Policy, including access, rectification, erasure, portability, objection, and withdrawal of consent. Requests specific to AI — for example deletion of your AI chat history or objection to a specific AI feature — can be made from in-app settings or by writing to [email protected].

11. Defensive scope — what AI Chef is not

ChefSphere's AI features are intentionally narrow. This section records the categories of regulated AI they do not fall into, so that the scope is unambiguous for regulators, researchers, and users.

11.1 Not a "companion chatbot" — California SB 243

AI Chef is a functional cooking, nutrition, and meal-planning assistant. It is not a "companion chatbot" within the meaning of California Civil Code §22601 (Senate Bill 243, effective 1 January 2026): it is not designed, marketed, or operated to meet users' social or emotional needs, does not exhibit an anthropomorphic persona, does not sustain social relationships across sessions, and is restricted by design to cooking, nutrition, meal-planning, budget, and recipe-extraction topics. Messages that attempt to use AI Chef for social companionship, mental-health support, or emotional-support functions are declined and, where safety-critical, re-routed to qualified-professional resources per §6.

As a matter of best practice — not statutory obligation — ChefSphere nonetheless implements the following SB 243-aligned safeguards on every AI Chef surface:

• Clear AI-identity notice on every conversation opener and on the first message of each session.
• In-app warning that AI output may not be suitable for some minors.
• Minor-user safety protocol that includes (i) a break reminder during extended continuous sessions, (ii) self-harm and suicidal-ideation detection with redirect to 988 (US), 112 (EU), Samaritans 116 123 (UK), (iii) a prohibition on romantic, flirtatious, or sexually suggestive output, and (iv) a prohibition on providing clinical diagnoses or treatment recommendations.

11.2 Not a "high-risk AI system" — EU AI Act Arts. 6 and 7, Annex III

None of the AI features described above fall within Annex III of Regulation (EU) 2024/1689 (the "AI Act") and none are prohibited under Art. 5. AI Chef, food-photo analysis, preference matching, recipe extraction, and meal-plan generation are transparency-obligation systems under Art. 50, not high-risk systems.

11.3 Not a "high-risk AI system" — Colorado AI Act (SB24-205)

AI Chef is not a "high-risk artificial intelligence system" within the meaning of C.R.S. §6-1-1701(9) (the Colorado AI Act, effective 1 February 2026). It does not make, and is not a substantial factor in making, consequential decisions about education, employment, financial or lending services, essential government services, health-care services, housing, insurance, or legal services.

11.4 Not a "deceptive synthetic media" operator — state deepfake laws

Where ChefSphere detects that user-submitted content is AI-generated or AI-modified, it applies the labelling rules in §8 regardless of the jurisdiction of origin. Deepfakes that impersonate real persons are prohibited by §8 and by the Acceptable Use Policy.

11.5 Not a "developer" of a generative AI system — California AB 2013

AI Chef and the other AI features described above are not "generative artificial intelligence systems developed by ChefSphere" within the meaning of California Assembly Bill 2013 (effective 1 January 2026, Cal. Bus. & Prof. Code §§22757.1–22757.5). ChefSphere is a deployer of third-party generative AI models via API — we do not train, fine-tune, substantially modify, or publish the underlying models. The AB 2013 training-data-transparency duty therefore rests with the upstream model providers listed by category in the Sub-Processor List §2, each of whom publishes (or is independently subject to) its own AB 2013 documentation. If ChefSphere ever trains, fine-tunes, or substantially modifies a generative AI model of its own that is made available to Californians, this Disclosure will be updated with the AB 2013 summary before the model goes live.

11.6 Google Play AI-Generated Content policy

ChefSphere's Android distribution complies with the Google Play AI-Generated Content policy: AI-generated content is labelled per §8 above; users can report offensive or illegal AI-generated content from every surface where such content can appear (in-app Report action + [email protected]); and the Android app surfaces the AI-feature disclosures required by the policy in-session and in the Play Store listing. The 15 April 2026 Google Play policy update (granular Health Connect permissions; Contacts Permissions policy) is reflected in the Android app manifest: ChefSphere does not request broad contact access, and health data collected via Health Connect is never used to determine employment or insurance eligibility.

12. AI literacy — AI Act Art. 4

Under Art. 4 of Regulation (EU) 2024/1689, providers and deployers of AI systems must ensure a sufficient level of AI literacy of their staff and of other persons operating or using AI systems on their behalf, taking into account their technical knowledge, experience, education, and the context in which the AI system is used.

• Internal literacy is ensured through onboarding modules, quarterly refresher training, an internal AI-risk register reviewed before material changes to AI features, and role-specific training for Trust & Safety reviewers who triage AI-assisted moderation signals.
• External literacy is supported through this Disclosure, in-app tooltips on every AI surface, the visible "AI-assisted" label on synthetic content, and the responsible-disclosure channel for AI-safety concerns at [email protected].

13. Changes

We update this document when AI features, providers, or applicable law change. Material changes are announced in-app or by email at least 30 days before they take effect, unless a shorter period is required by law or by a safety concern.

14. Contact

• Data Protection Officer: [email protected]
• Privacy: [email protected]
• AI safety / responsible disclosure: [email protected]