Carli Michelle Heller, a resident of Bonita Springs, Florida, was taken into custody after her interactions with Anthropic’s AI chatbot, Claude, contained statements about planning a shooting at the Lee County Sheriff’s Office. The messages, dated September 26, indicated an intent to use a firearm, and a follow-up entry on September 27 mentioned that she had acquired a new gun. Law enforcement acted on the information and arrested Heller without incident, charging her with a written threat of violence.
Anthropic's safety protocols
Anthropic’s consumer agreement grants the company discretion to share user input with authorities when it believes a serious risk exists. The firm’s safety system scans conversations for specific trigger phrases and potentially harmful content. When the algorithm flags a message, it is escalated to a human review team for further assessment. In Heller’s case, the reviewers determined the threat warranted police notification, prompting the subsequent investigation.
Legal framework and company policy
The company’s privacy policy, updated in September 2026, permits disclosure of data to law-enforcement agencies when there is a good-faith belief that such a step is necessary to prevent serious harm. Even users who opt out of data use for model training still have their chats retained for up to three months, allowing the company to retain visibility over flagged content. The policy also states that flagged exchanges may be incorporated into future training to improve the system’s detection capabilities.
Prior similar incidents
Anthropic has previously reported user threats to authorities. In September, the firm alerted police after a user claimed to have purchased an AR-15 and expressed hostile intent toward the company’s CEO. Although that individual was not immediately detained, Anthropic cited the episode as evidence that its safeguards were functioning as intended. Comparable procedures exist at other AI firms; for example, OpenAI’s policy routes threatening content to human reviewers who may involve law enforcement when the risk is deemed credible.
Potential ramifications
Heller’s court appearance is scheduled for November, where she will face a felony charge for the written threat. The case highlights the growing intersection between AI platforms, user privacy, and public safety responsibilities. Companies that host conversational AI must balance the promise of private, personal interactions with obligations to intervene when users express violent intentions.
Why it matters
The incident underscores how AI providers are increasingly positioned as first responders to digital threats, shaping the legal landscape around data privacy and corporate liability. As more users treat chatbots as personal journals, the expectation of confidentiality may clash with safety protocols that compel companies to involve law enforcement. This tension will likely influence future regulatory discussions on the responsibilities of AI service providers in monitoring and reporting harmful content.
