Could well be a GTF that is not sorted in the correct order. I don't know how the cufflinks parser works, but for many GTF parsers, duplicate IDs are detected by finding that exons the same ID are not all next to each other. This can happen if the file is sorted purely on chromosome and start position as transcripts from the same gene will be interleaved and overlapping genes will be interleaved. You get around this you would need to sort by gene_id, then transcript_id, then start position.