Researchers tested several AI models using a classic psychology exercise called the Stroop task. The test is designed to measure attention and self-control in humans.

In the task, people see color words like “red” or “blue” written in different ink colors and must name the ink color instead of reading the word. This becomes harder when the word and color do not match.

The AI systems performed well on short lists of words, but their accuracy dropped sharply as the lists became longer. Some models went from over 90 percent accuracy to below 20 percent in longer tests.

When matching and conflicting words were mixed together, performance dropped even more. Researchers say the systems often returned to simply reading the words instead of following the instruction.

The study suggests that unlike humans, who can usually control attention and ignore distractions, AI systems still struggle with this kind of mental control, especially in longer and more complex tasks.

The findings highlight that despite impressive abilities, AI still has limits when it comes to focus and resisting distraction.