AI tools could turn social media posts into convincing phishing emails, study finds

A few photos from a concert can reveal more than you might expect. Researchers at the University of Texas at Arlington found that widely available AI tools can use public social media posts to create phishing emails tailored to someone’s recent experiences. The messages may sound believable because they refer to details the recipient actually shared.

The approach, known as context-aware spear phishing, uses information about a person’s everyday life to make a scam feel personal. For example, someone who posted photos from a trip might receive a message that appears to come from their hotel, offering a discount on a future stay. The link, however, could lead to a fake site designed to steal login or payment details.

“Anyone can be the attacker, and that makes it scarier,” said Shirin Nilizadeh, an associate professor of computer science and engineering at UTA who led the research. Public posts may give a scammer enough information to craft a convincing message without needing special knowledge of the person.

The research team studied 200 public Instagram accounts and found that as few as 10 posts could provide enough material to create targeted emails. The researchers assessed more than 18,000 AI-generated messages, comparing them with thousands of real phishing emails. The AI drew on details such as travel photos, upcoming events and concerns people had shared online. It could also customize subject lines, sender names and links.

A post about an upcoming concert, for instance, might prompt a fake ticket confirmation or an offer to upgrade seats. Someone who mentioned a delayed flight could later receive a fraudulent message offering compensation. In a user study, participants generally considered the AI-generated emails more convincing and less suspicious than the real-world examples used for comparison.

There is an important limitation: the study measured people’s reactions in a controlled setting, not how often they would click malicious links during an actual attack. Still, the results suggest that familiar details can make it harder to recognize a scam.

Scammers may not need to break into an account to gather those details. If a profile is public, its posts can provide a useful record of someone’s interests, plans and recent activities. A message that arrives at just the right time—and mentions something that really happened—may feel more trustworthy than a generic phishing attempt.

AI can also make these scams easier to produce at scale. Traditional warning signs, such as spelling errors or awkward wording, may be absent from polished AI-generated messages. An email can sound professional and refer to a real event in someone’s life while still being designed to steal their information.

The researchers also tested three AI safety tools: Llama Guard, ShieldGemma and WildGuard. With their default settings, the systems identified only 0% to 19% of requests to create personalized phishing messages. After the researchers adjusted the safeguards to look specifically for context-aware phishing, detection rates reached as high as 99% in their tests.

Those results suggest that stronger protections are possible, though they do not guarantee that every malicious request would be stopped outside a research setting. The team did not publish the exact instructions it used to generate the phishing emails, to avoid giving criminals a ready-made template. Nilizadeh said AI companies should strengthen safeguards against targeted social engineering.

You don’t have to stop sharing photos, but it may help to review who can see your posts. Consider limiting access to approved followers and checking older posts for details you no longer want public. Even private posts can be screenshotted or shared by someone who has access.

Sharing vacation photos after you return, rather than while you are away, can also avoid revealing your location in real time. Be mindful of location tags and posts announcing upcoming plans. Small choices like these may make it harder for strangers to track your activities.

If a message from a hotel, airline or other business refers to something you recently did, verify it independently before responding. Open the company’s official app or type its website address into your browser instead of clicking the email link. You can also call a number you have confirmed separately. Be wary of urgent requests to sign in or provide payment information; sender addresses can be spoofed.

Keep your devices and security software updated, and use the available protections against suspicious websites and attachments. These tools can help, but they cannot catch every phishing attempt—particularly if you enter your details on a convincing fake website. If an email looks suspicious, report it through your email provider’s phishing-reporting option.

Personal details may also appear on data broker and people-search websites, where scammers could find information such as phone numbers or home addresses. You can contact those sites to request removal, though doing so may take time and cannot erase everything available online.

Use a different, strong password for each important account. A password manager can help you create and store them. Turn on multifactor authentication where available, preferably using an authenticator app or security key, and never give a verification code to someone who contacts you unexpectedly. Passkeys can provide additional protection against phishing on services that support them.

Clicking a phishing link does not automatically mean your device has been compromised, but it is worth taking suspicious messages seriously. When an email seems unusually well-timed or includes personal details, pause and check it through another channel. That extra moment could help prevent a costly mistake.