Theory of mind is commonly assessed with the false belief task, which involves attributing a false belief to a character. Children with Autism Spectrum Disorders are said to perform less well on this test than typically developing children or children with intellectual disabilities. This might suggest a deficit in theory of mind in autism, but this idea is currently being challenged. Other skills seems to be involved in the false-belief task, such as language pragmatics, the ability to interpret language in context. Indeed, the test question is ambiguous: it can lead to different interpretations. Failure could therefore be due to a difference in interpretation between the child and the experimenter. For people with limited pragmatic language skills (such as people with autism or typically developing children under 4), the pragmatic ambiguity of the test question could be an obstacle to success, even if they are otherwise able to attribute a false belief to others. Nevertheless, changing the context could improve performance: the test question, usually asked by an adult experimenter, could become less ambiguous when asked by a NAO robot presented as an ignorant student (mentor-child paradigm). 26 children with autism (and 24 children with intellectual disabilities) passed the false-belief task with a human or with a robot, in order to assess whether they perform better in this new context. The results show no improvement in performance with this paradigm, but children with ASD may have been distracted by the robot during the test, which would have diminished their success.