Should you upload your DNA to ChatGPT? What happens to a 23andMe or Ancestry file

The Morse teamPublished October 30, 202610 min read

If you upload DNA to ChatGPT, you hand OpenAI a copy of your genotype that, as of October 30, 2026, is saved to your Library, outlives the chat you used it in, and can train OpenAI's models unless you turn off Improve the model for everyone. Claude and Gemini keep uploads on their own schedules too, and HIPAA, the law that protects your medical records, doesn't reach a chatbot you use on your own. The file also says things about your parents, siblings, and children, who never agreed to share it.

This isn't medical advice, and it isn't an interpretation of anyone's DNA. It's about where a 23andMe or Ancestry raw data file goes once you ask a chatbot to read it. We read 23andMe's and Ancestry's own help pages and 23andMe's privacy statement, its bankruptcy filings and sale announcement, the upload and retention pages of OpenAI, Anthropic, and Google, the NIH genome institute's pages on genetic privacy and discrimination, HHS's page on who HIPAA covers, and the California and Illinois genetic privacy statutes, all on October 30, 2026. 23andMe and Ancestry deserve credit for one thing. Both warn you, in plain words, that your download leaves their protection.

What's in a 23andMe or Ancestry raw data file?

Your genotype, one marker per line. 23andMe says the file "consists of lines of your genotype call data (your A's, T's, C's and G's)" (23andMe), and that each line gives the marker's name, its exact location in the genome, the possible variants, and the ones found in your saliva (23andMe). It arrives as a zipped text file whose name begins with "genome," usually about a week after you ask for it, and sometimes up to 30 days (23andMe). 23andMe has described its data as covering "hundreds of thousands" of these markers (23andMe, 2015).

Ancestry's file has the same shape: a marker ID, the chromosome, the position, and your two letters at that spot, delivered as dna-data-(date).zip through an emailed link that works once and expires after a week (Ancestry). Ancestry says the same data is used "to determine your ancestral origins, match you with genetic relatives, and predict your traits."

Both companies say the copy you download is on you. Ancestry: "Once you download your DNA Data, your downloaded copy is no longer under the protection of our security measures." 23andMe: "23andMe cannot secure Personal Information that you release on your own," and "third parties may use your personal information differently than we do."

Can ChatGPT read my DNA raw data file?

It can open it. ChatGPT takes text files up to 512 MB, capped at 2 million tokens per file (OpenAI). Claude accepts TXT and CSV files up to 500 MB in a chat, but ZIP isn't on its list, so you'd unzip the file first (Anthropic).

Whether the reading is any good is a separate question, and 23andMe answers part of it. It says its raw data has had "a general quality review," but "only a subset of markers have been individually validated for accuracy," and that the data is "suitable only for informational use and not for medical, diagnostic, or other use." A chatbot reading the file inherits those limits.

What happens to a DNA file you upload to ChatGPT, Claude, or Gemini?

Each company keeps it differently. Personal accounts, read October 30, 2026.

Anthropic gives its periods for chats rather than for files separately, so a file in a Claude chat goes when the chat does. Every app on this list has one more way back in, feedback. OpenAI says that if you press thumbs up or down, "the entire conversation associated with that feedback may be used to train OpenAI models," even after you opt out (OpenAI). Anthropic stores "the entire related conversation" for up to 5 years when you press thumbs up or down, and Google keeps feedback, "any included content (like files)," for up to 3 years. For the rest of each company's settings, see does ChatGPT train on your data, is Claude private, and is Gemini private.

Is it safe to upload DNA to ChatGPT with training turned off?

Turning training off doesn't remove the file. Gene Food's guide to uploading genetic data to ChatGPT, updated September 15, 2026, points readers to the account setting that keeps data "not used to train the model" (Gene Food). That setting changes training only. OpenAI's Library page says "Deleting a chat containing files does not delete a separate file that remains saved in Library," and its upload page adds that "Content from a file that was brought into a conversation can remain in that conversation" (OpenAI). So a raw data file has two copies to clean up: the file in Library, and whatever ChatGPT quoted from it in the chat.

The same guide tells readers to look for platforms that sign HIPAA Business Associate Agreements. A consumer can't get one: OpenAI says Health in ChatGPT "does not offer a Business Associate Agreement" (OpenAI).

Does HIPAA or GINA protect DNA you give a chatbot?

Not HIPAA. HHS says "the HIPAA Rules apply to covered entities and business associates," meaning health care providers, health plans, clearinghouses, and the contractors they hire, and that an entity outside those definitions "does not have to comply with the HIPAA Rules" (HHS). We go through where that line falls for chatbots in private AI for health questions.

GINA, the Genetic Information Nondiscrimination Act of 2008, limits what two kinds of company can do. Health insurers may not use genetic information for eligibility, coverage, or premiums, and employers may not use it for hiring, firing, or pay (NHGRI). The genome institute lists what it leaves out:

• Other insurance. GINA's protections "do not cover long-term care insurance, life insurance, or disability insurance," though some states add their own.
• Small employers. GINA "does not apply to employers with fewer than 15 employees."
• The military, which "is permitted to use genetic information to make employment decisions."

GINA is a rule about what insurers and employers may do with genetic information. It says nothing about what a chatbot company keeps.

State law goes further in places. California's Genetic Information Privacy Act defines a "direct-to-consumer genetic testing company" to include any entity that "analyzes genetic data obtained from a consumer" or uses genetic data "directly provided by a consumer" (California Civil Code 56.18). Whether that reaches a general chatbot reading an uploaded file isn't something the statute says by name, and we found no court or agency page that settles it. Illinois's Genetic Information Privacy Act bars a direct-to-consumer testing company from sharing your genetic test information with a health or life insurer without your written consent, and lets people sue for $2,500 per negligent violation or $15,000 per intentional or reckless one (Illinois General Assembly). This isn't legal advice either.

What happened to 23andMe's data after the bankruptcy?

It was sold with the company. 23andMe filed for Chapter 11 on March 23, 2025, in the Eastern District of Missouri, and on June 27, 2025, the court approved the sale of substantially all its assets to TTAM Research Institute, a California nonprofit led by co-founder Anne Wojcicki, for $305 million (SEC filing). TTAM announced on July 14, 2025 that it had completed the purchase and that 23andMe keeps operating under it (23andMe). Two days before the filing, California's Attorney General urged customers to consider deleting their data and having their samples destroyed (California Attorney General).

23andMe's privacy statement, last updated May 19, 2026, says your information "may be accessed, sold or transferred" in a bankruptcy, merger, or asset sale. It limits genetic data to a buyer that is a US nonprofit research institution, or a company that satisfies the Committee on Foreign Investment in the United States, and agrees to adopt the statement. It also says 23andMe and its lab keep your genetic information, date of birth, and sex for legal compliance "even if you chose to delete your account" (23andMe).

None of the upload pages we read from OpenAI, Anthropic, or Google sets a rule for genetic data. A genotype you upload follows the general rules in the table above, like any other text file.

Does uploading your DNA expose your relatives?

Yes, because you share DNA with them. Ancestry uses the data to "match you with genetic relatives," and 23andMe's breach shows what that sharing can do. In its December 1, 2023 SEC filing, 23andMe said an attacker got into "a very small percentage (0.1%)" of accounts with reused passwords, then used them to reach "a significant number of files containing profile information about other users' ancestry" shared through its DNA Relatives feature (SEC filing).

The genome institute says that, except for identical twins, each person's DNA is unique, "which means a DNA sample can never be truly anonymized" (NHGRI). It cites a 2013 study that re-identified research participants by pairing genomic data with genealogy databases and public records. Removing your name from a chat doesn't make the genotype in it anonymous, for you or for the relatives it describes.

How do you ask AI about a DNA result with less on record?

Send the question, not the genome. A question about one marker needs that marker's line, or the result written out in words, and it doesn't need the other hundreds of thousands of lines or your name. If you use ChatGPT, a Temporary Chat keeps the file out of Library, and Gemini's temporary chats are kept for 72 hours. Don't rate the reply. If you've already uploaded a file, delete it from ChatGPT's Library as well as deleting the chat.

Our guide to local AI vs cloud AI covers what keeping data on your own computer protects, and what it doesn't. For the rest of the threat model, see how to use AI privately.

How Morse keeps your DNA questions private

Every row of the table above ends with a copy of your genotype on a company's servers, filed under your account, under a policy the company can change or a buyer can inherit. 23andMe's own statement shows how far a sale can carry data. Morse is built so your chats don't sit on our servers in the first place.

• E2EE, for a result you'd show no one else. Paste the few lines your question is about, or describe the result in words. Your browser checks the secure enclave's hardware attestation before anything leaves your device, then encrypts the message so only that enclave can read it. Our servers relay ciphertext they can't read, and the reply is decrypted only in your browser. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed.
• Private, for general genetics questions you'd rather no host keep. New chats start here. Open-weight models like DeepSeek, Kimi, and GLM, plus Grok, answer under Morse's zero data retention agreements with its inference, search, and other providers, so they keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
• Anonymous, to ask GPT, Claude, or Gemini without the provider knowing it's you. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider still sees the text and may keep it for a limited time to check for abuse, under its own policy. A genotype identifies you by itself, so paste only the lines the question needs.

No chat or file waits on our servers to be cleaned up later. A file you attach is sent with the message and never stored by Morse, and your chat history is stored in your browser, encrypted, not on our servers. Morse doesn't train AI on your chats, sell your data, or show ads, so there's no training switch to find. Because we don't store chats, we usually have very little to give authorities (privacy policy). Clearing your browser or losing the device deletes your chats for good, and Morse can't recover them, so keep the original file and your test company's reports where they are.

Try a chat from the Morse home page without an account, or start free with no card needed. End-to-end encrypted chats come with Pro, at $19 a month.