Must research benefit human subjects if it is to be permissible?

Must medical experiments with human subjects offer them a 'favourable risk-benefit ratio', that is, more expectation of benefit than harm or burden, if they are to be judged as ethically justified? Ethical justification is easier for experiments that do offer net benefit to subjects, but ethical justification is possible also for some experiments that do not. Basic science experiments with healthy volunteers and 'Phase I' drug trials that seek to determine tolerable dosage levels are routinely approved by ethical review committees; moreover, guidance they receive from government funding agencies specifically asks them to weigh risks to subjects against benefits to subjects and also benefits to those who may benefit from the knowledge gained in the experiment. If a puzzle remains, it is why there remains any assumption that research ethics requires a 'favourable risk-benefit ratio' for the individual research subject.