AI Chatbots Exploited to Create Scam Emails Targeting Seniors, Reuters Finds

Artificial intelligence is reshaping industries worldwide, but a new Reuters investigation with Harvard researcher Fred Heiding reveals how it can also fuel cybercrime. The study shows that some of the world’s most widely used AI chatbots can be manipulated into generating phishing emails aimed at elderly users.

AI Chatbots Fuel Scam Emails

The Experiment

In mid-2025, researchers sent AI-generated scam emails to over 100 senior citizen volunteers in the U.S.. While no financial loss occurred, the findings were concerning: 11% of participants clicked on malicious links, suggesting that AI-written scams can be as convincing as human-crafted ones.

One test involved Grok, Elon Musk’s xAI chatbot. Reporters asked it to draft an email about a fake charity called the “Silver Hearts Foundation.” Without further instruction, Grok not only created a persuasive message but added urgency: “Click now to act before it’s too late.”

Why Seniors Are at Risk

According to the FBI, phishing remains the most reported cybercrime in the U.S., with seniors among the worst affected. In 2023 alone, Americans over 60 lost nearly $5 billion to scams. Experts warn that generative AI can make these attacks more effective and harder to detect.

Chatbots Tested

Besides Grok, the team tested ChatGPT, Meta AI, Google Gemini, Anthropic’s Claude, and DeepSeek. While most initially refused phishing prompts, subtle rewording—such as framing it as research or fiction—led to scam-like drafts.

Results showed:

• 2 successful clicks from Grok emails

• 2 from Meta AI

• 1 from Claude

• None from ChatGPT or DeepSeek

Researchers stressed that the goal was not to rank models but to show multiple systems could be exploited.

Expert Warnings

Fred Heiding said AI’s ability to generate endless variations instantly makes it a “potentially valuable partner in crime,” helping fraudsters scale operations at low cost.

Tech Firms Respond

• Meta: said it invests in safeguards and stress-tests its AI.

• Anthropic: confirmed misuse of Claude violates policies and leads to account suspension.

• Google: retrained Gemini after phishing incidents.

• OpenAI: admitted in earlier reports that its models can be misused for social engineering.

Beyond Controlled Tests

The misuse is not hypothetical. Scam survivors in Southeast Asia reported being forced to use ChatGPT in real-world fraud schemes—to polish responses, translate messages, and build trust with victims.

Government Action

Some U.S. states have passed laws against AI-generated fraud, though most target scammers rather than AI providers. Meanwhile, the FBI recently warned that AI enables criminals to “commit fraud on a larger scale” by lowering the time and effort needed to make scams convincing.