Americans ask AI chatbots about almost everything, from health to relationships to big existential questions. While those conversations might feel enlightening or intimate, they are not always one thing: private.
A Washington Post review of public records and local news stories found a dozen instances of chatbot transcripts being cited in the public record in court cases over the past two years. In many cases, the transcripts were unearthed from users’ devices when searched by a police officer or an opposing party during the evidence-gathering stage of a civil case. In a handful of instances, AI companies have reported disturbing content on their platforms to law enforcement.
“Unless you are having a chat with a service that has a temporary chat or, basically, an incognito version … [and] you’re also having it within a browser that’s not tracking you, the answer is no. You can’t be sure that it’ll be totally private,” said Jen King, a privacy researcher at the Stanford Institute for Human-Centered Artificial Intelligence.
Here’s what to know about the privacy of your chatbot conversations.
Where do my chatbot conversations go?
When you use an AI chatbot, such as Claude or ChatGPT, your message is sent to a data center or other company server that processes your query and returns a response. If you’re logged into an account, you can go back to your phone or computer later to view your chat history.
In court cases reviewed by The Post, chatbot transcripts were often retrieved from searches of users’ devices. In one case, a Missouri man agreed to let an officer search his phone without a warrant, according to police records. In another in Michigan, a defendant’s lawyer volunteered a search of his client’s devices as a show of transparency, according to court records.
If you delete a conversation from your account, both OpenAI and Anthropic — the makers of ChatGPT and Claude, respectively — say they will delete it from their internal servers within 30 days in most cases. (The Post has a content partnership with OpenAI.)
Exactly how long chatbot companies retain your queries on their servers depends on the type of account you use. On regular, personal accounts, OpenAI will keep your chats until you delete them. The company also offers temporary chats that are automatically deleted within 30 days. (Anthropic, likewise, calls them “incognito chats.”) But there are exceptions; OpenAI says it may retain deleted data, including temporary chats, if your account has been banned for violating its policies or if the company is legally required to do so.
Can my employer see my chatbot conversations?
Many workplaces have opted into enterprise agreements with OpenAI and Anthropic that prevent the AI companies from training models on their employees’ inputs, which could include proprietary information. In general, the companies’ terms of use and privacy policy for consumer accounts may not necessarily apply to enterprise agreements.
But don’t assume that your activity on your company account is private. Just like email or Slack, a corporate Claude or ChatGPT account is managed by your employer. Anthropic, for instance, offers tools for some administrators to see information such as chats and files.
Both companies offer zero data retention for select enterprise customers. But that option is not available to consumer accounts.
Can AI companies contact authorities about my conversations?
OpenAI has faced criticism after deadly shootings allegedly committed by people who were heavy users of the bot. OpenAI had banned the account of the suspect in a February mass shooting in Tumbler Ridge, British Columbia, for violating its policies but declined to report the conversations to Canadian authorities. CEO Sam Altman later apologized for not alerting law enforcement. OpenAI has said it has a “zero-tolerance policy” for use of its tools for violence.
OpenAI says it uses automated detection systems to identify potentially concerning activity, such as violence or self-harm. The company says it notifies law enforcement when “conversations indicate an imminent and credible risk of harm to others.” Anthropic, meanwhile, says it will disclose limited information if it has a “good-faith belief” that disclosure is necessary to prevent serious harm to people or property.
Both companies say they comply with valid subpoenas.
What’s the best way to keep my chats private?
It depends on what aspects of privacy are most important to you. Both OpenAI and Anthropic say they do not share conversations or sell data to third-party advertisers. If you want to ask something you don’t want your employer to see, it’s probably best to keep that out of your workplace account.
Deleting a chat or using temporary ones might mean your usage won’t be stored by an AI company indefinitely. There are certain kinds of personal information you might want to avoid telling AI altogether, such as your Social Security number or passwords.
Privacy is hard to come by online. In messaging, the gold standard for security is the Signal app, which has an auto-delete feature and encrypts traffic end-to-end as it travels through the company’s servers, meaning that even Signal cannot see the content of your messages.
There’s no analog for AI — and most apps, search engines and other software aren’t nearly as privacy-intensive as Signal. Just as even incognito browser searches get routed through Google, OpenAI and Anthropic still have to see your question to answer it.
The post Your boss, tech companies and police can read your chatbot conversations appeared first on Washington Post.
