In a recent study conducted by the University of Texas at Arlington, computer‑science researchers demonstrated how artificial‑intelligence tools can scrape publicly posted photos and updates from social‑media platforms and automatically craft convincing phishing emails. The findings highlight a growing threat to ordinary users who share everyday moments online.

How AI creates “context‑aware” phishing

The team, led by associate professor Shirin Nilizadeh, examined 200 public Instagram accounts. They discovered that as few as ten posts—such as concert photos, travel snapshots, or a complaint about a delayed flight—provided enough context for AI to generate a targeted email that appears to come from a trusted source, like a hotel, airline, or event venue.

In a controlled user study, participants rated the AI‑generated messages as more believable and less suspicious than real‑world phishing examples. While the researchers did not measure actual click‑through rates in a live attack, the results suggest that scammers could exploit familiar details to lower a recipient’s guard.

Scale and automation raise alarm

The study evaluated more than 18,000 AI‑generated emails, comparing them with thousands of authentic phishing messages. The AI was able to customize subject lines, sender names, and links to match recent personal experiences, making the scams feel unusually personal.

Unlike traditional phishing, which often relies on generic language and obvious spelling errors, AI‑crafted emails can sound professional and polished, eliminating many of the visual cues users have learned to spot.

Current AI safety tools fall short

To test existing safeguards, the researchers submitted phishing‑generation requests to three popular AI safety systems—Llama Guard, ShieldGemma, and WildGuard. With default settings, detection rates ranged from 0 % to 19 %, indicating that many malicious prompts could slip through.

When the team adjusted the safety prompts to specifically flag “context‑aware phishing,” detection improved dramatically, reaching as high as 99 % in their laboratory tests. This suggests that stronger, more targeted safeguards are possible, though they remain unproven in everyday deployments.

Practical steps for everyday users

While the research does not call for abandoning social media, it does urge users to be more thoughtful about privacy settings. Making accounts private or limiting the audience for posts can reduce the amount of personal data available to potential attackers.

Other recommended precautions include:

• Verifying the sender’s email address before clicking any links.
• Hovering over URLs to see the actual destination.
• Using two‑factor authentication on email and financial accounts.
• Keeping security software up to date.

By taking these simple steps, individuals can add layers of protection against increasingly sophisticated AI‑driven scams.

Implications for policymakers and tech companies

Nilizadeh and her colleagues stress that AI developers must prioritize robust safety mechanisms before releasing powerful language models to the public. The study’s findings provide concrete evidence that without stronger safeguards, AI could become a tool for mass‑scale social engineering.

As AI continues to integrate into everyday applications, both industry leaders and legislators will need to consider regulations that balance innovation with the protection of personal privacy and security.

Original reporting: Fox News (HLL/CB) — read the source article.