This is a probe (more properly called a directory harvest attack, DHA), which merely seeks valid email addresses to send future scams to. (Invalid email addresses, or those that SMTP-reject based on spam convictions, will bounce. See also listwashing.)
vrfy command, which was designed to allow testing addresses without sending mail, is disabled on almost every SMTP server because its overwhelming (if not exclusive) use is for directory harvest attacks (see RFC 5322 §7.3: VRFY, EXPN, and Security). On such systems, spammers need to actually send actual emails in order to test addresses.
This also tests the sending infrastructure. Minimal message content and no payload URL help evade anti-spam and therefore lower the odds of SMTP rejection based on content (which would generate a bounce and offer one less potential victim for the actual attack).
It could additionally serve as a poisoning attack, aspiring to weaken receiving anti-spam systems so a followup spam with an actual payload has a better chance at landing in your inbox.
Typical hash busting and Bayesian poisoning attacks try to morph message content to avoid detection, often by adding gibberish. I'm talking about a less direct variant, where the goal is harming the training and tuning of these systems to they are weakened so that a future attack might get through.
An empty body message means an ML system like Bayesian anti-spam needs to rely on the tokens it collects from the message headers. ML struggles to handle these messages because there is so little content (especially when the spam comes from a shared infrastructure like a free service like Yahoo or GMail). A fuzzy hashing system would have to draw data from headers or else use a dangerous template of the body content.
In an effort to adjust and catch this sort of spam, some systems may end up generating false positives. As a result, the signal gets worse and that anti-spam component may get a lower weight.
Some systems, such as SpamAssassin's TxRep and AWL plugins, will bias the spam score of a known sender's mail toward the average of their past mail. They could therefore reduce the actual spam payload's apparent spaminess.
Of course, another option is that somebody didn't configure their spam templates correctly 😉