The freely available sequencing and clinical data are the types of data that have a "very small" risk of being used to identify an individual. Anything that poses more than a "very small" risk of identifying an individual has controlled access.
Determining this risk of identification has a substantial statistical basis. This link shows an outline of how such assessments are made.
In general, anything that could mapped to an individual's normal genomic DNA sequence has restricted access. For example, gene-expression results from RNAseq data are freely available with respect to the genes that were matched, but the actual sequences matched from an individual are protected as they could include germ-line mutations specific to an individual.
Identifying a fusion gene typically requires analysis of whole-genome sequencing data or finding fusion-gene products in RNAseq data. In either case, there would have to be access to the specific individual sequence data as in FASTQ or BAM files.
It is possible to be granted access to the controlled-access files, provided that those responsible for guarding the data are satisfied that there is a valid scientific purpose and that the potentially identifiable data will be protected by the user. The repositories describe their processes for getting such access; see this page for the NIH dbGaP that controls TCGA data.