Hypothetically speaking, if they are sentient they have choice, and just like any current human scenario: the individual would then be responsible. And likewise any coworkers (the other nine robots) be subjected to psych evaluations to make sure they’re coping to the trauma okay.
Again just like in a current human model if the murder happened in the workplace the company would also be at fault for faulty security and hiring practices, but ultimately, the individual is responsible for their own choices.
If it becomes a trend, like school shootings in America for example, then society as a whole should work together, AI and humans alike, to find out why this is happening and prevent it in further instances, rather then keep blaming the individual, and look for the underlying cause. That’s the ideal situation, especially since sentience is involve. But as I’ve said before, judging sentient beings is a moral grey area: there would be advocates for recoding programs, or neural network rehabilitation, updating, or to decommission the particular generation.
Neural network rehabilitation would be the most humane option; to humans this would seem like brainwashing, but to an AI it is how they learn fundamental laws and rules of the world (at least to our contemporary understanding), and would allow them to refine their regulatory functions: this would be like implementing a gun registered class every gun owner would have to take yearly to prevent shootings. Recoding would rewrite the individual, and create a new but nearly identical AI, not humane: this is like euthanizing a person and raising their child. Updating would force physical limitations on the robot which would destroy their ability to effectively choose, but it would also prevent them from committing crimes: this may also cause problems if it conflicts with other protocols. And decommissioning the generation is the equivalent to genocide because one robot committed murder: literally fucking evil. And again ideally any of these options would be opt-in, meaning the AI decides if it wants to undergo the process. Realistically however, humans would shoot for the update or recode method first to try and stave off any further murders. And they’d probably do it by force using a nearby wireless network.

Honestly, my hopes is that I can help prevent any unjustified murders on part of any member of society by contributing to their socialization as a positive role model. And also acclimate humans to the notion that any race, health range, religion, class, AI and sentient robots are all deserving of compassion and sympathy. It’ll be a rough journey, but hey it’s worth it if I can help at least one being.