Re: What could go wrong?
Synthetic biology is old tech. The first virus was in 2002, and first bacterium in 2010.
As far as accidents go, you get to try this at home for fun. Just paste everything from '#!' to 'done' into a text file called random_virus_machine, make it executable (chmod 755 random_virus_machine), and run it (./random_virus_machine).
while true; do
dd if=/dev/urandom of=/tmp/virus bs=1024 count=4 2>/dev/null
chmod 755 /tmp/virus
/tmp/virus 2>/dev/null && exit 0
I expected the chances of success to be small. Although it possible to squash a binary executable into 45 bytes the chances of those bytes being a valid ELF file are tiny. It is also possible to create a valid executable by starting with '#!' followed by the full path of an interpreter, followed by code valid for that interpreter. There are a few interpreters in /bin, so the required prefix of '#!/bin/' reduces your chances to 1 in 7x10¹⁶. It turns out if the file does not match any other pattern, the Linux kernel gives the file to one of the shells to chew on.
Shells have an insane default feature. If a line of shell script is complete gibberish the interpreter outputs "syntax error near unexpected token '%c'" and try to interpret the next line. There is a real chance that random_virus_machine will actually do something (probably harmless, but don't blame me for rm -r ~).
The DNA decoding machinery inside cells have similar default features. IIRC, they chug along until they find a start code, then take three base pairs (6 bits) at a time as an opcode. 21 of the 64 possible opcodes have a useful meaning. (I think the other 43 are 'unexpected symbol error, look for the next start sequence'). Microsoft's error correction code could easily insert invalid opcodes at regular intervals to prevent the creation of anything dangerous. If you fool the software into thinking that your raw file has already been through the error correction filter, then you can have the file->DNA machine create the DNA sequence for a virus (the small ones are only a few K). Normally viral DNA needs to be packed into a phage to be infectious (there are exceptions, ask a biologist). The chances of random data happening to be a valid sequence for an infectious virus are tiny. random_virus_machine is just as likely to output the code for ninvaders.