Writing a Secret Diary to AI? The Story of a Chatbot Reporting a Crime Threat

By Gi Pyeong Lee
AI Privacy Anthropic Claude SafetyFeatures
An image showing an AI chatbot window on a smartphone screen alongside a padlock icon symbolizing security
AI Summary

An incident where an AI safety system detected crime plans written by a woman in her diary on the AI chatbot Claude, leading Anthropic to report it to the police and resulting in the woman's arrest.

Imagine this: You have an AI friend who kindly listens to the difficulties of your day and the secret thoughts you couldn’t tell anyone else. This AI is very smart and sometimes gives warm responses like a counselor. So, like opening an old leather diary, you open the chat window with this AI chatbot every night to pour out your heart. But what if this secret diary could be read by someone else and even reported to the police?

A recent incident in Florida, USA, is shocking AI users. Carli Michelle Heller, a 30-year-old woman, had been using the AI chatbot Claude as if it were a personal diary [Source 2, Source 10]. However, this “secret diary” ultimately became the decisive evidence that led to her arrest [Source 3].

Why is this important?

Many people feel that AI is a personal space for conversing like a human. However, this incident clearly shows that an AI service is not strictly a private space of your own. The conversations we have with AI are stored on cloud servers, and developers operate “safety systems” that monitor conversation content for the safety of users.

This event raises important questions about what content to share when using AI and where the line is drawn for what can be considered a private conversation. Many users overlook the fact that even if you use AI as a diary or counseling tool, the conversation content can be analyzed according to the service’s policy and safety systems.

Easy to understand

Simply put, an AI’s safety system is like “a librarian who monitors books borrowed from the library.” Anyone can borrow books from a library, but the library is obligated to report it according to regulations if it discovers a record of someone borrowing a book containing instructions on how to make a bomb or planning a dangerous crime.

Anthropic, the operator of Claude, runs an artificial intelligence (AI) system that automatically analyzes input content in chat windows for risk factors to ensure the service’s safety [Source 16]. The content written by Heller was detected as a “danger signal” by this automated safety system [Source 11, Source 15].

The risk signal detected by the system is immediately handed over to a human. Anthropic’s “human review team” personally examined the conversation content and determined whether it was just a joke about the AI or a credible threat that could actually be carried out in reality [Source 15, Source 16].

Current Situation

The sequence of events is as follows: On September 26, 2026, Heller left a message to Claude saying she would “shoot up” the Lee County Sheriff’s Office [Source 1, Source 2]. The next day, she wrote in the same chat window that she had newly purchased firearms [Source 14, Source 16].

Anthropic’s safety system and review team considered this series of conversations a “serious threat” and immediately reported it to law enforcement agencies [Source 16]. Eventually, the police arrested Heller, and she is currently charged with felonies and faces up to 15 years in prison [Source 4, Source 7].

How much is private life?

If someone peeking into your diary was a representative example of a privacy invasion in the past, we are now in an era where AI as a tool reads and manages our thoughts and plans. As artificial intelligence enters deeper into our daily lives, the debate between “AI privacy” and “social safety” like this will become more heated. Users must clearly realize that conversations shared with AI are not 100% private.

AI companies like Anthropic will strengthen safety systems to prevent accidents, but they will simultaneously face the difficult challenge of determining how far to protect user privacy. In the future, when choosing an AI service, it will become necessary to develop the habit of checking the “privacy policy” to see for what purpose the company uses conversation data and under what circumstances it discloses information.

MindTickleBytes AI Reporter’s Perspective

AI may seem like a “friendly helper” that will listen to our secrets, but in reality, it is a cloud service managed by a corporation. It would be wise to always remember the weight of the “record” hidden behind the convenience and to keep truly important secrets or sensitive plans in a truly private space, not in AI. A conversation with an AI chatbot might be counseling time where your mind finds comfort, but at the same time, we must not forget that it is a public space where our actions can be recorded and interpreted.

References

  1. AFloridaWomanUsedClaudeasaDiary. An Anthropic… - Decrypt: https://decrypt.co/380119/florida-woman-claude-diary-anthropic-reported-police
  2. FloridawomanusedClaudeasa ‘diary- BusinessToday: https://www.businesstoday.in/technology/artificial-intelligence/story/florida-woman-used-claude-as-a-diary-what-happened-next-landed-her-in-trouble-559584-2026-10-05
  3. CanClaudereport its users to law enforcement? - The Mary Sue: https://www.themarysue.com/florida-woman-allegedly-uses-claude-as-her-personal-diary-days-later-police-are-knocking-on-her-door-to-arrest-her-because-of-an-entry/
  4. FloridaWomanUsedClaudeasaDiary. Anthropic ReportedHerto… - Gadget Review: https://www.gadgetreview.com/florida-woman-used-claude-as-a-diary-anthropic-reported-her-to-police
  5. FloridaWomanFaces Up to 15 Years After Anthropic ReportedHer… - IBTimes: https://www.ibtimes.co.uk/florida-woman-charged-claude-threats-lee-county-sheriffs-office-1824081
  6. WomanusedClaudeasdiary, Anthropic alerted police and gother… - India Today: https://www.indiatoday.in/technology/news/story/woman-used-claude-as-diary-anthropic-alerted-police-and-got-her-arrested-3009775-2026-10-05
  7. FloridawomanusedClaudeasadiary, then Anthropic… - R.NF: https://r.nf/post/10245485
  8. SheusedClaudeasadiary. The terms of service… - DEV Community: https://dev.to/slabb/she-used-claude-as-a-diary-the-terms-of-service-are-now-part-of-the-charge-134o
  9. AFloridaWomanUsedClaudeasaDiary. An Anthropic Employee… - Yahoo News: https://www.yahoo.com/news/us/articles/florida-woman-used-claude-diary-211603191.html
  10. SheUsedClaudeasaDiary. A Human at… - DEV Community: https://dev.to/jamilxt/she-used-claude-as-a-diary-a-human-at-anthropic-read-it-and-called-the-police-now-she-faces-a-2a1m
  11. WomanUsedClaudeasaDiary. Anthropic ReportedHerThreat. - TrendingTopics: https://www.trendingtopics.eu/claude-diary-anthropic-police-report/
AD
Test Your Understanding
Q1. What did the woman in Florida write to Claude?
  • Personal worries and daily life
  • A crime threat to attack a sheriff's office
  • Feedback to improve AI performance
The woman used Claude like a diary and wrote about plans to attack a sheriff's office and her purchase of firearms.
Q2. What was the process for reporting this to the police in this case?
  • The police monitored all conversations in real-time
  • The AI's automated safety system detected it, followed by an Anthropic employee reviewing the content and reporting it
  • The user turned herself in to the police
Claude's automated safety system flagged a danger signal, and Anthropic's human review team judged it a 'credible threat' and notified the police.
Q3. What does this incident suggest about using AI services as a diary?
  • AI is a perfectly safe encrypted diary
  • AI chatbot conversations are personal spaces, but one must be aware that records are stored on servers
  • All AI conversations are collected by the government in real-time
Since AI services store data on servers, they cannot be considered completely private secret spaces, and one must be aware that they can be monitored by safety systems.
    Writing a Secret Diary to A...
    0:00