Memes are self-replicating constructs that "care" mostly about their own spread. They are an analogy to genes, only that they are purely informational. Like genes, memes don't necessarily have their hosts best interest in mind and can sometimes even destroy the host. While a too destructive meme would, like a gene causing a child to be stillborn, not be very successful evolutionarily speaking, it can nonetheless exist in the design space of all possible memes.
Assume that my research group has proven that a meme, a perfectly convincing argument for suicide, can in fact exist. We now want to engineer this memetic weapon. Obviously it would have to be a multi-stage weapon, as it could not be deployed in a save manner otherwise.
While it is obvious and feasible (see the Manhatten Project) to run such a project by letting most people operate on a need to know basis only, there is usually a group at the center that fully understands the whole project.
Considering the nature of the weapon, we know that there exists a population of memes S in the space of all possible memes, which will cause a person to make committing suicide their top priority after they comprehend the gist of the meme. This is not about magical words, this is about certain complexes of believes/ideas. The exact nature of any of these memes is not yet known and will be explored over the course of the research project.
As a sidenote, there exist a number of other meme populations with other and in some cases even naturally occurring behavioural consequences. These could also be explored as part of the project.
The issue is, how do we find and optimise the weapon without killing ourselves? Assume that our government has no issues providing us with disposable subjects who can be made to cooperate and an inexhaustible source of funds
Addendum 1: I am aware of this question, but it is about the general feasibility/existence of these memes, not about the logistics of creating one.
Addendum 2: I am aware of the existence of mild memetic killing agents (for example the blue whale game), but this isn't what I am after. Most memes can in some way damage the host in a certain context (just consider on the genetic side of things that it a possibility cause of aging that certain genes are beneficial in early life, yet detrimental in later life). What I am after is not a generic weapon of limited mass destruction, which can spread far and wide and can take out a low percentage of its hosts. I'm more interested in something like the killing joke, which was mentioned in the comments. Near 100 percent lethality, as this makes it more reliable and it self-contains it's spread.