One item is driving this story. SitePoint has published a piece titled "Claude's Steganographic Request Marking: What Developers Need to Know," which surfaced through Google News's Anthropic feed.
That headline is, at present, the entirety of the sourced material. There is no technical specification, no statement from Anthropic, no stated launch date, and no outside comment available in the source items — so anything beyond the title should be treated as unconfirmed.
What the title does tell us is the shape of the thing. According to SitePoint's framing, the feature is called steganographic request marking, it lives in Claude, and the intended audience is developers rather than general users.
Steganography, broadly, is the practice of hiding information inside other information — a message tucked into a file or a piece of text in a way that isn't obvious to someone reading it normally. "Request marking" implies that requests are being tagged in some fashion. How those two ideas combine in Claude's implementation is not described in the material available here.
The gap is itself worth noting, because the label alone raises the questions developers would most want answered: what exactly gets marked, whether the marking is detectable or removable, who is able to read it, and whether it changes anything about how applications built on Claude should be designed or audited.
Until Anthropic publishes documentation — or the SitePoint article's specifics are independently confirmed — the responsible read is that a feature with a notable name exists and that the details are still outstanding.
It matters because anything that quietly marks AI requests touches trust, traceability, and privacy at once, and developers building on Claude will need to know which of those it is before they can judge it.