Collaborative research holds great promise. Not only would our disciplines benefit from it, but it would also represent a definite added value for the actors involved: collective learning at the very least, recognition and emancipation at times. As long as the ambition remains modest and lucid, I occasionally practice a group analysis method that assumes these objectives and promises (Van Campenhoudt, Chaumont & Franssen, 2005). However, I believe that this methodology can keep its promises because researchers remain masters of the game and its cognitive aims would be threatened if, under the guise of research ethics in particular, this power were to be abandoned to the actors. I will take as a case study a research project in which I was involved in a "partnership" between my university and the ATD Fourth World Movement between 1996 and 1998. This research is now given by ATD as an example of a "rigorous methodology that has been tried and tested for years" which "would have triggered a kind of 'big-bang' since a growing number of initiatives […] are developing this method or are inspired by it". After having shown the deficiency, both scientific and political, of its results, I will explain it by the renunciation of the elementary requirements of any reflexive approach. I will then give other examples where researchers have been asked by associations to renounce all or part of their conclusions for the price of their collaboration. Finally, I will conclude on the need to carefully preserve the autonomy of the scientific field even and especially in collaborative research.