AI Disclaimer

WhisperBee uses AI to (a) turn your voice notes into searchable text, (b) extract structured fields like queen status or brood pattern from those notes, (c) summarize each hive into a daily briefing, and (d) answer questions about a hive in plain language.

All four can be wrong. They can transcribe the wrong word, infer the wrong queen status, miss something obvious, or invent something that wasn’t in your audio. We design the UI to keep you in the loop — every AI-extracted field is editable, every briefing is just a starting point, and every chat answer is grounded in the data you’ve actually logged.

If a hive looks sick, a queen is failing, you suspect a notifiable disease (AFB, EFB, varroa thresholds your state defines), or you’re about to make a decision that costs colonies — call your state apiary inspector, your veterinarian, or your local extension agent. WhisperBee is not a substitute for them.

There are four AI surfaces in the product. Each does one narrow thing:

WhisperBee is not:

• A veterinary diagnostic device. It cannot diagnose American Foulbrood, European Foulbrood, Nosema, viral disease, varroa-vectored DWV, or any other condition. If you suspect a notifiable disease, contact your state apiary inspector — most states require it by law.
• A regulated medical or agricultural product. WhisperBee has no FDA, EPA, or USDA registration, and its outputs are not approved as labels, prescriptions, treatment protocols, or pesticide guidance.
• A substitute for your state’s required record-keeping. Some states and programs (e.g., USDA ELAP loss documentation) impose specific record formats and retention rules. WhisperBee aims to make the underlying data exportable, but it’s on you to confirm the export meets your jurisdiction’s requirements.
• A safety-of-life system. Don’t use AI suggestions to make time-critical decisions about anaphylaxis, sting reactions, or any human-medical situation. Call emergency services.
• Personalized to your operation in any way that overrides general beekeeping best practice. If your local mentor or extension agent says one thing and WhisperBee says another, follow the mentor.

Specific failure modes we’ve seen in our own testing — call them out so you know what to watch:

• Mishearing beekeeping jargon. “Apivar” can come out as “a pivar.” “Supersedure” sometimes becomes “super seed-er.” Re-read the transcript before saving if the extraction looks off.
• Inferring queen status from incomplete notes. If you said “saw eggs but not the queen,” the AI might mark the queen as “present” (correctly inferring laying activity) or as “not seen” (correctly noting you didn’t spot her). Both can be right. You decide which.
• Mite load interpretation. A description like “a few mites on the sticky board” will not be converted into a percentage threshold. The AI extracts what you said; it does not infer alarm levels you didn’t state. Always do an alcohol-wash or sugar-roll if you need a number.
• Stale briefings. A briefing reflects the last time it was generated. If you log new data and don’t refresh, the briefing is out of date. The UI shows a “stale” chip when this happens.
• Hallucinated treatments or events. Rare but possible. If a structured field appears that you don’t recognize from your audio, edit it. Don’t accept it on trust.
• Seasonal context errors. WhisperBee maps your latitude to a band (tropical / subtropical / temperate / northern / subarctic) and uses that to set seasonal expectations. If your microclimate doesn’t match your band, suggestions can be off by weeks.

Specific situations where you should ignore WhisperBee output and contact a qualified human:

• Suspected notifiable disease (AFB, EFB, exotic pests). Contact your state apiary inspector or extension service. Most states legally require reporting; WhisperBee’s output is not a substitute.
• Treatment dosing. Always follow the product label. Labels are regulated; WhisperBee guidance is not. If WhisperBee’s suggestion conflicts with the label, the label wins, every time.
• Prescription medications (e.g., terramycin under VCPR rules that apply to honey bees in the US). Talk to a veterinarian. WhisperBee is not a Veterinary-Client-Patient Relationship and cannot create one.
• Hive movement across state or international lines. Inspection certificates, permits, and quarantine rules vary. Use the right authority, not the app.
• Dead colony forensics. A WhisperBee briefing might guess “mite collapse” from your last inspection notes, but the definitive diagnosis comes from an extension lab or the USDA Bee Research Lab via the National Honey Bee Disease Survey.

WhisperBee’s AI features are powered by two third-party providers:

More detail on what data leaves your device is in our Privacy Policy, Sections 4 and 5.

Under our data processing agreements with both OpenAI and Anthropic, your voice recordings, transcripts, and inspection content are not used to train, fine-tune, or improve their models. They are processed solely to return a result for you, then handled under each provider’s standard retention policy (typically 30 days for abuse-monitoring purposes; nothing longer).

WhisperBee likewise does not use your hive data to train any model — neither our own nor a third party’s.

• Edit any AI-generated field before saving. Your edit is what gets persisted; the AI’s original guess is overwritten.
• Manual-entry mode for inspections — skip the AI entirely and type your inspection by hand. Available on every inspection screen.
• Delete any voice recording from the inspection detail screen at any time. The audio file and the corresponding row in our storage are removed.
• Delete your account from Settings. All AI-generated outputs tied to your account are purged synchronously; subprocessor copies expire under their own retention windows (typically 30 days).
• Daily AI quotas. Each tier has a per-day cap on transcription and Claude calls. If you hit the cap you’ll see a 429 / over-quota message; you can still log inspections in manual-entry mode and AI processing will resume the next day.
• Limit Sensitive Personal Information use (California residents) — see Privacy Policy §8.3.

During the WhisperBee beta, AI behavior is being actively tuned. Expect:

• Occasional changes to extraction confidence thresholds, briefing tone, and action-item rule firing. We may release these without prior notice.
• Higher false-positive rate on health flags than the eventual production target (we’d rather flag too much during beta than too little).
• Sentry-collected error reports tied to your account so we can debug AI regressions. These reports do not contain your voice or note content (see Privacy Policy §2.5).

We’ll bump the “Last updated” date on this page when the AI behavior changes meaningfully.

If WhisperBee’s AI gives you a bad output that affected (or could have affected) a hive decision, we want to hear about it. Email support@whisperbee.app with:

• What the AI said
• What was actually correct
• The inspection date or hive ID, if you have it handy
• A screenshot if it’s a UI surface (briefing, chat answer)

We use these reports to (a) tune extraction prompts, (b) add suppression rules for false positives, and (c) decide whether the underlying issue belongs in our AI evaluation suite as a permanent regression test.

Related reading:

• Terms of Service §5 — legally binding AI disclaimer
• Privacy Policy §4 — AI processing data flow
• Beta Tester Addendum