Original title: Johnny Still Receives Spam SMS: Assessing the Robustness of SMS Spam Detection
Muhammad Salman, Muhammad Islam, Muhammad Ikram, Mohamed Ali Kaafar · September 2026
Phone spam filters look great in tests, but clever spammers still slip past them in real life. Attackers alter junk texts just enough to trick the filter while keeping the message readable. The researchers tested real tools people rely on, including commercial messaging apps, third-party services, and public models. They attacked them using noticeable tweaks and sneaky, invisible changes that actually survive real SMS or RCS text delivery. They found major security gaps across existing spam detectors. Teaching a model on one style of trick only helped it stop similar tricks. The defense failed when faced with entirely new, hidden encoding tactics. To fix this, the team built a team of varied models. This ensembleCombining multiple models to make a final decision together. blends adversarial trainingTraining a model on tweaked examples to make it more resistant to attacks. with detectors that use different architectures and text-processing methods. Voting together, this group model caught both obvious and hidden attacks far better, while still accurately spotting normal text messages.
Imagine a security guard at a party checking guest lists. Normally, they spot fake invitations by obvious errors. But if someone creates a near-perfect fake with tiny adjustments, the guard might miss it. Here, the spam detectors are like the security guards, and the tweaked messages are the clever fakes. The paper shows that even after training guards on known fakes, new kinds of fakes still slip through, so they recommend using a team of guards with different strengths to catch them better.
If spam systems can't handle these tricks, more unwanted messages reach your phone, increasing risks of scams and privacy breaches for everyone who uses text messaging.
The abstract does not specify the exact datasets or models tested in detail, so it's unclear how the results generalize to all real-world scenarios. The evaluation focuses on SMS and RCS delivery, not other messaging platforms, and the ensemble method's effectiveness may depend on the diversity of models used.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Johnny Still Receives Spam SMS: Assessing the Robustness of SMS Spam Detection — Muhammad Salman, Muhammad Islam, Muhammad Ikram et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.