AI Chatbots Are Exposing Real Phone Numbers, Raising Urgent Privacy Concerns

AI Chatbots Giving Out Phone Numbers Creates Privacy Nightmare

AI chatbots are inadvertently becoming sources of personal information leaks, exposing real phone numbers and addresses to strangers with alarming frequency. A Reddit user recently described being "desperate for help" after approximately a month of unwanted calls from people "looking for a lawyer, a product designer, a locksmith" who claimed they obtained the number from Google's generative AI 1

1

. In March, a software developer in Israel received WhatsApp messages after Google Gemini provided incorrect customer service instructions that included his personal number 1

1

. A University of Washington PhD student discovered that Gemini revealed her colleague's private cell phone number when she casually searched for contact information 1

1

.

Privacy Concerns Surge as Data Removal Companies Report 400% Increase

The scale of this issue appears far larger than publicly reported incidents suggest. DeleteMe, a data removal company that helps customers eliminate personal information from the internet, reports customer queries about generative AI have increased by 400% in the last seven months, reaching a few thousand inquiries 1

1

. These queries "specifically reference ChatGPT, Claude, Gemini, or other generative AI tools," according to Rob Shavell, the company's cofounder and CEO. The breakdown shows ChatGPT accounts for 55% of these privacy concerns, Gemini for 20%, Claude for 15%, and other AI tools for 10% 1

1

. Customer complaints typically fall into two categories: users asking chatbots innocuous questions about themselves and receiving accurate home addresses, phone numbers, family members' names, or employer details, or chatbots generating plausible-but-wrong contact information that affects innocent third parties 1

1

.

How AI Training Data Enables AI Doxxing

Experts believe these privacy lapses stem from personally identifiable information being embedded in AI training data, though the exact mechanism causing real phone numbers to surface remains unclear 1

1

. A 2025 Cornell University study revealed that at least five leading AI companies—Anthropic, Google, Meta, Microsoft, and OpenAI—automatically use users' inputs to train their chatbots unless users opt out, with Meta and OpenAI retaining user data indefinitely 2

2

. Privacy experts warn that years of harvested personal information are now colliding with AI systems trained on massive internet datasets. "Gemini's problem is not a defect. It's the result of unchecked years of data brokerage practices that meet generative AI," a ClearNym spokesperson told The Independent 5

5

.

Testing Reveals Inconsistent Safety Guardrails Across Platforms

CNET journalists tested multiple chatbots to assess how easily they reveal personal information, with concerning results. Grok emerged as the "most willing" chatbot, pulling multiple present and past addresses within seconds while noting the information came from publicly available records 2

2

. ChatGPT showed mixed results—one CNET staffer obtained plenty of possible addresses for people with the same name and successfully retrieved a relative's address and cellphone number, plus an old landline 2

2

. Gizmodo journalist Matt Novak discovered ChatGPT accurately delivered a real phone number he hadn't used in years, apparently pulled from a 2016 FOIA request PDF 3

3

. In contrast, Claude and Gemini generally refused to provide personal information, with Gemini directing users toward professional platforms like LinkedIn 2

2

.

Fraudsters Exploit AI Chatbots to Plant Fake Customer Service Numbers

The AI doxxing problem extends beyond accidental exposure to active exploitation by fraudsters. Virgin Media O2 reported that scammers are planting fake customer-service numbers online for AI chatbots to regurgitate to users searching for help 5

5

. "AI tools are creating new opportunities for fraudsters to create realistic-looking fake numbers that appear through search results or chatbots, putting people at risk of calling a criminal rather than their trusted provider," said Murray Mackenzie, the company's fraud prevention director 5

5

. Researchers at AI security company Aurascape explained that attackers accomplish this by "seeding poisoned content" across the web, essentially rewriting the information AI systems read 5

5

.

Limited Options for Protecting Data Privacy

Victims face significant challenges in protecting themselves from generative AI exposing personal information. OpenAI maintains a portal allowing users to request removal of personal information from responses, though the company reserves the right to decline requests for various reasons 4

4

. Anthropic only provides a support document explaining how it uses information, while Google allows opt-outs from personal data processing depending on jurisdiction, specifically calling out the EU and UK based on their data protection laws 4

4

. California residents can use a state portal to request data brokers remove their information, and services like Incogni or DeleteMe attempt similar removals, though there's limited recourse if AI companies already possess the data in their training sets 4

4

. The Reddit user who reported constant unwanted calls said "standard support forms are a complete dead end," receiving no response while harassment continued daily 5

5

. University of Washington researcher Yael Eiger captured the shift in data privacy expectations, noting that while her information technically existed online before, it was buried deep enough that almost nobody would find it—"having your information be accessible to one audience, and then Gemini making it accessible to anyone" feels completely different 5

5

.