Catch of the Day: AI Phish
Examples of clever phish that made it past my anti-spam nets and into my inbox. Some are contributed by clients or readers like you, and other reliable sources on the Internet.
You can send phishing samples to me at phish@wyzguys.com.
My intention is to provide a warning and show current examples of phishing scams, related articles, and education about how these scams and exploits work, and how to detect them in your inbox. If the pictures are too small or extend off the page, double clicking the image will display them in a photo viewer app.
Analysis of Phishing Emails Shows High Likelihood They Were Written by AI
It’s no longer theoretical; phishing attacks and email scams are leveraging AI-generated content based on testing with anti-AI content solutions.
I’ve been telling you since the advent of ChatGPT’s public availability that we’d see AI’s misuse to craft compelling and business-level email content. According to cybersecurity vendor Abnormal Security’s latest report, “AI Unleashed: 5 Real-World Email Attacks Likely Generated by AI in 2023,” the likelihood of emails today being written by AI is very high.
According to the report, “the Abnormal platform determines the probability that an attack was AI-generated by utilizing ChatGPT—a newly-launched tool that leverages a suite of open-source large language models (LLMs) to analyze how likely it is that a generative AI model created the message.”
They provide some color-coded examples, like the one below to make the case.
The color coding indicates just how predictable the next word would be based on its context to the left. The green words represent that each one was one of the top 10 predicted words to come from AI. Yellow words are from the top 100 most predicted words, and red, the top 1000.
The green coloring is pretty damning, as it’s unlikely that a human would be so predictable as to match LLM-based AI tools. The report also notes the lack of typos and grammatical errors, meaning that these emails will be more compelling and more difficult to identify as being written by a hacker (because they weren’t).
While we’re just talking about text content, there’s no reason why an AI tool can’t be trained to look at legitimate email content from known brands to craft near-perfect emails that will fool just about anyone.
Blog post with color-coded screenshot:
https://blog.knowbe4.com/analysis-ai-generated-phishing-email-content
Share
JAN
About the Author:
I am a cybersecurity and IT instructor, cybersecurity analyst, pen-tester, trainer, and speaker. I am an owner of the WyzCo Group Inc. In addition to consulting on security products and services, I also conduct security audits, compliance audits, vulnerability assessments and penetration tests. I also teach Cybersecurity Awareness Training classes. I work as an information technology and cybersecurity instructor for several training and certification organizations. I have worked in corporate, military, government, and workforce development training environments I am a frequent speaker at professional conferences such as the Minnesota Bloggers Conference, Secure360 Security Conference in 2016, 2017, 2018, 2019, the (ISC)2 World Congress 2016, and the ISSA International Conference 2017, and many local community organizations, including Chambers of Commerce, SCORE, and several school districts. I have been blogging on cybersecurity since 2006 at http://wyzguyscybersecurity.com