Operator: Nickelsense Inc. (dba Parachute, Tether, Tandem)
Applies to: Parachute, Tether, and Tandem websites at https://myparachute.live, https://mytether.live, and https://mytandem.live, the Parachute, Tether, and Tandem apps for iOS and Android, and related services
Effective: October 2, 2026
Last updated: October 2, 2026
Version: 1.0
This Notice explains the optional choice to let Nickelsense Inc. use selected information to train and improve AI models. It is separate from the Terms, product analytics, recovery research, and marketing, and from ordinary use of an AI feature. It supplements Section 14 of the Privacy and Consumer Health Data Policy.
Using an AI feature to answer your current request does not give us permission to train on your information. Training a reusable model on your information requires the separate opt-in below. Choosing No does not reduce your access to core features.
What we use without a separate opt-in. We continuously use aggregate engagement statistics — for example, how often features are used and which kinds of content and actions members engage with — to operate the Service and to improve and train our systems, including our AI models. These statistics contain no name, contact information, or account identifier, no text you write, and no health or recovery answers, and they are analyzed only in aggregate.
"Help improve Nickelsense AI?" If you choose Yes, Nickelsense may use only the categories you check to train, fine-tune, evaluate, and improve Nickelsense AI models (models we host ourselves and models we fine-tune through our AI providers) for the purposes in Section 3. Every box starts unchecked:
Never eligible, regardless of consent: direct messages, journals, crisis and safety reports, moderation reports, credential documents, payment data, precise location, contact lists or hashes, voiceprints or raw voice, and information about another person.
| Model owner | Nickelsense Inc. |
|---|---|
| Model name and version | Nickelsense AI models (the model version in use is shown in the App when you opt in) |
| Purpose | Improve recovery-resource recommendations, detect unsafe AI output, and improve the tone and helpfulness of AI peer-support responses |
| Who can use the trained model | Nickelsense brands (Parachute, Tether, Tandem) and white-label installations operated by Nickelsense |
| Commercialization | Not sold or licensed separately; made available only as part of the Service, including white-label installations |
We and our processors may use selected data only to prepare a controlled training corpus, train or fine-tune, evaluate safety, quality, and bias, and audit the named model for the named purpose. A materially different model, recipient, or purpose requires a new choice. If you opt in, you grant Nickelsense a nonexclusive, worldwide, royalty-free license to use the selected content for the stated training purpose during the consent period. It does not transfer ownership and does not authorize advertising, sale of your source content, identification of you, or unrelated model training.
Before training, we confirm you are 18 or older and record your consent version; remove direct identifiers and unnecessary metadata; exclude named third parties and high-risk content; generalize or suppress dates, location, rare attributes, and small groups; keep the corpus key separate from account systems; prohibit re-identification and provider secondary use by contract; test for memorization, extraction, membership inference, and disparate performance; and apply the HIPAA de-identification standards (Safe Harbor or Expert Determination) as an engineering benchmark. Deidentified does not mean risk-free: rare combinations or model memorization can create residual risk.
Each AI provider we use for training (Anthropic, OpenAI, Google, or xAI, as applicable) processes only the deidentified training examples you selected in the United States for the duration of the training job, and no longer than our contract allows. A provider may not use the data to train its own or other customers' models, or for advertising, sale, profiling, or re-identification, and must delete source data at the end of the contract.
Source examples stay linked to your consent record for 24 months or until you withdraw, whichever comes first, then are deleted. Training datasets and checkpoints are kept for 24 months. Model versions are kept as long as the model version is in use, plus 12 months for security and audit purposes.
Withdraw at any time at Settings → Privacy → AI training. Withdrawal stops new collection and future training use linked to you; removes linked source examples from active training sets within 30 days; causes us to instruct processors to delete them; and triggers retraining, machine unlearning, or retirement of the affected model, or a documented risk review where those steps are not technically feasible, where required. A model that has already been trained may not let us isolate one person's influence, so we do not promise individual "model deletion" unless it is technically supported. We will tell you what we did and any lawful exception we relied on.
You may ask which categories you contributed, correct the source where applicable, withdraw, request deletion, and complain or appeal at Settings → Privacy (in the App or on the web) or privacy@nickelsense.com. Your consumer-health-data rights under the Privacy Policy still apply.
These boxes start unchecked:
Buttons: "Allow selected AI training uses" / "No thanks." We log the notice version, selected categories, model and purpose, providers, timestamp, locale, and any withdrawal and remediation.
Other legal documents
Nickelsense Inc. · Louisville, Kentucky · legal@nickelsense.com · (502) 354-8105