The following is my original answer, which I stand by as being logical and on the face of it applicable, though it does not seem to tell the whole story. See below for a revised answer.
You are right: it should be who here.
This is a simple (and fairly common) case of hyper-correction, where people who are not quite sure how to use who and whom correctly end up using whom in places where it should not be used because they think it sounds ‘fancier’ and therefore more formal (a logical enough thing to think, since whom is indeed mostly used in formal contexts). This happens even in places where whom is quite clearly impossible to anyone who distinguishes the two forms, such as “I did not know the person whom was sitting next to me”.
In this instance, however, the mistake is easier to understand, even if made by someone who distinguishes who and whom: the intervening parenthetical phrase “it is foreseeable” probably influenced the choice. It is obvious that it in that parenthetical is the subject, which probably made the author think who(m) would then have to be an object—having two subjects right next to each other is not common and usually avoided.
Revised answer that tells a fuller story (based on comments by Araucaria and myself below, some of which have been incorporated into the answer itself and therefore deleted to improve the legibility of the comment chain)
As Araucaria has shown below, there appears to be more to this than meets the eye. This particular construction is not as straightforward as an initial analysis that simply removes the “it is foreseeable” as parenthetical information might suggest, for the following reasons:
There are many instances of the oblique-case pronoun whom being used in this type of construction, especially in the field of law, but also in literature and elsewhere
The deletion properties of the pronoun in these constructions match those normally found with object pronouns, rather than subject pronouns
For examples of (1) above, I refer to the quotes unearthed by Araucaria in the comments below and in the comment section of the question itself. They show quite clearly that, whatever prescriptive (or even descriptive) grammar says, forms with whom are quite common and widespread, even in proofread, formal texts written by able language users.
If you transpose various elements in the sentence, you really end up with a fairly complex structure.
… persons whom it is foreseeable are X …
– is made up of a head noun and a following relative clause. The relative clause itself, however, is made up of a main clause and a complement clause (in layman’s terms often called a that-clause). The pronoun (which, being a relative pronoun in a relative clause, must stand at the head of the clause) is really the subject of the complement clause embedded in the main clause. In other words, if we split up the clauses to avoid a relative clause, strange things happen. Compare:
I have two friends who are fans of Lady Gaga. => I have two friends. They are fans of Lady Gaga.
– where the relative clause is the same as the resulting main clause, except for the pronoun; with:
I have two friends who(m) I think are fans of Lady Gaga. => I have two friends. I think (that) they are fans of Lady Gaga.
– where the complement clause (that) they are fans of Lady Gaga is split up to allow the pronoun to be raised to the head, enabling the containing main clause to function as a relative clause. The relativiser is extrapolated from an embedded clause, rather than being present in the main clause itself. The only thing that could be relativised in the main clause (it => which) is left alone for semantic reasons: it is not a possible anaphor for the antecedent persons
With this much moving around, it is hardly surprising, really, that extrapolation of an embedded subject to become the relativiser in the containing clause can wreak some grammatical-logical havoc. Even with this analysis in mind, I still cannot determine what grammatical role the extrapolated pronoun ought to play in the relative clause (as opposed to in the embedded complement clause, where it is clearly the subject). It seems, if we go by actual usage, that it can be interpreted as either subject or object.
One fact that proves that the interpretation as object is real and not just people being confused about when to use whom is relativiser deletion. In a relative clause, the relativiser1 must be present if it is the subject in the clause:
My friend who loves Lady Gaga.
– but is optional if it plays a non-subject role in the clause:
The Lady Gaga song (that) my friend loves most. (object)
The Lady Gaga album (that) my friend gave a 10/10 rating. (indirect object)
The Lady Gaga dress (that) my friend wrote a school essay about. (prepositional object)
In the example from this question, the relativiser can be deleted, which clearly shows that it is interpretable as a non-subject element (my guess would be it’s probably some kind of vaguely adverbial element more than a real object):
[Y]ou owe a duty to persons (who[m]) it is foreseeable are X …
If we consider “it is foreseeable” to be a parenthetical statement and remove it, this deletion is not possible:
*[Y]ou owe a duty to persons (∅) are X …
This suggests that the relativiser in your sentence (with “it is foreseeable”) plays the role of non-subject, and whom is appropriate.
On the other hand, if we look closer at the complement clause itself, a very striking and rather annoyingly contradictory phenomenon can be gleaned.
Normally, in English complement clauses, the complementiser (that) is optional. In some cases, it sounds clunky to have it, in others it sounds clunky to leave it out—but it is usually not ungrammatical to do either. This is also the case in constructions like the present ones, but only if the relativiser is a non-subject in the complement clause. Regardless of what role the (also optional) relativiser who(m)/which/that plays in the relative clause (= main clause + complement clause), the complementiser that can be left out or kept in as long as the relativiser is not the subject in the complement clause:
The Lady Gaga album (which) I think (that) my friend is listening to.
Here, both relativiser which and complementiser that can be left out or kept in with no change of any kind.
If the relativiser is the subject of the complement clause, however, the complementiser appears to be mandatorily blocked—it simply cannot appear:
The Lady Gaga album (which) I think (*that) is playing now.
This is most peculiar, and I have absolutely no idea why it is so; but it is. Having the complementiser in the sentence here is completely and utterly ungrammatical in any register or form of English I’ve ever seen.
In your sentence here, adding a complementiser makes the sentence quite ungrammatical:
Persons (who) it is foreseeable (*that) are likely to…
This suggests that the relativiser must be playing the role of subject, and who is appropriate.
This is rather messy, but it seems to be true: the relativiser here is either both subject and object or neither subject nor object, depending on your point of view. No wonder people get confused about whether to use who or whom!
1 For present purposes, I consider it unnecessary to distinguish between complementisers and operators (=relativisers), since they are mutually exclusive anyway. I thus use relativiser for both that and who(m).
Best Answer
I don't think that there is a relevant difference between sentences like "He is demanding £5,000 from the elderly woman whom he claims has ruined his life" and "Whom should I say is calling?" If there is a relevant difference, it would probably be that "whom" is a relative pronoun in the first, and an interrogative pronoun in the second. Possibly, interrogative whom is less common than relative whom, but I haven't seen much literature that discusses the evidence for that idea in detail.
The word whom is fronted in both sentences. The 2012 Language Log post "Sometimes there's no unitary rule," by Geoff Pullum, refers to "a preposed relative or interrogative who" when discussing this topic.
However, JK2's answer points out that in Chapter 5 of the CaGEL (published in 2002) the use of whom in this context is described as being mostly restricted to relative clauses. I don't know whether Pullum's views have evolved since the publishing of the CaGEL, or whether the inclusion of interrogatives in the rules given in the 2012 blog post was an accidental oversight.
There is a 2004 blog post by Pullum ("I really don't care whom") that describes the general state of the who vs. whom distinction as confusing and says that whom "hardly occurs in interrogatives at all".
Whom would not be an "object" here
It's not correct to categorize whom as "the object of the whole clause", any more than he is the object in sentences like I know that he is calling. Rather, it is "not the subject" of the relative clause. Hopefully the distinction I'm making between "not the subject" and "object" is understandable, if nitpicking.
Whom is "incorrect" here from a prescriptivist perspective
The sources that you have seen are all taking the prescriptivist viewpoint. The prescriptivist rule isn't inherently invalid; the point that descriptivist linguists like Pullum are trying to make in documents like the linked blog post is that the pattern of usage that the prescriptivists condemn has a logic of its own, and it seems likely that many people who write things like "the person whom I say won the race" aren't just accidentally using "whom" where they say "who" (i.e. it's not just a typo or "production error"), and aren't just consciously choosing to write "whom" because they think it is more formal and they don't know where it should be used (that kind of thing is what linguists like Pullum would call a "hypercorrection"), but rather are following a rule that they have internalized to some degree that says to use "whom" in contexts like this. So it is "grammatical" for them in the sense that it is consistent with a rule that they plausibly have acquired as part of their own personal system of grammar.
From a descriptive viewpoint, things that are consistent with a speaker's grammar are not described as "mistakes" or "errors". "Acceptability" depends on the grammar of the listener/reader. The articles that you have seen use words like "mistake", "wrong", "incorrect", "error" in a different sense, to refer to things that aren't consistent with the rule that prescriptivists have historically preferred. For example, from a prescriptive viewpoint things like "ain't" or "haven't got no" are "incorrect". This is a common use of these words in popular discussions of grammar, but modern linguists tend to avoid using these words this way in scientific contexts.