I think they’d be more amiable the more research is put into them. Ideally, multiple uses would be allowed. However, even the an AI might build up a tolerance to the behaviors of the virus; thus requiring different or not malicious viruses to feel the same buzz.

Edit: to clarify, now that I can type on the comfort of my desktop keyboard.  Simulated drugs run their own risks.

1. The problem being one must be able to code the virus to cease function after a duration of time.  Or else the effects of the drugs may become committed to long term.  Unless the drugs effect is to shut down the pathways between long term and short term memory.  Which would cause the equivalent of brain damage after prolonged use.
2. If you found a way to kill the drug program, the next problem becomes preventing addiction.  Just as with chemical drugs, a digital drug might seem more favorable than reality.  This may cause an artificial dependency on the drug program.
3. Preventing the drug virus from mutating.  Viruses often steal or change data, it’s implied in it’s name.  This means that a drug may make itself more or less potent, change it’s behavior, or embed itself in code, making it harder to treat or deviate.  Making a malware that inherently alters AI code isn’t just hacking the network within the brain, it affects every aspect about how or even if the AI functions.
4. Just like physical drugs, it could change their personality if not coded properly.
5. Just like chemical drugs, it could kill them, if not coded properly.

Thanks to @worldmeetsrobot for inspiring this addition.