Zero. That’s how many legitimate copyright offices will email you asking to verify your registration. Yet authors are reporting exactly those messages right now, sent by someone impersonating the United States Copyright Office. That’s the state of the writer’s inbox as authors push back against claims from publishers and agents over the Anthropic settlement, and as scammers circle the same waters.
I build bots for a living, so my instinct when I see this pattern isn’t outrage, it’s architecture. Money is moving. Claims are being filed. Ownership is contested. Every one of those conditions creates a window where a stranger with a convincing email signature can insert themselves into a transaction. Writer Beware has been tracking this shift for years: direct solicitation became one of the main ways scammers recruit victims, alongside ads. The other side of that coin is that direct solicitation is also how plenty of real business gets done, which is precisely why it works so well as a disguise.
The verification problem is a routing problem
If you’ve ever written a message handler, you already understand what’s happening to authors. An untrusted message arrives. It claims an identity. It requests an action, usually one that’s irreversible. In software we’d never let that through without checks. In email, most people read the display name and move on.
The fix is boring and effective: treat inbound claims like untrusted input. Here’s the mental pipeline I’d wire up if I were an author fielding messages about rights, claims, or representation.
- Identity check first, content check second. Does the sending domain match the organization it claims to represent? Not the display name, the actual domain after the @. Scam messages lean on lookalike domains and free mailboxes.
- Out-of-band confirmation. Never reply to verify. Look up the organization independently and contact it through a channel you found yourself. A reply-to address is attacker-controlled data.
- Direction of money. Legitimate representation does not typically ask you to pay upfront to receive what you’re owed. Any message where the money flows toward the sender deserves the highest scrutiny.
- Urgency as a signal. Deadline pressure is a social engineering primitive. When a message manufactures time pressure, that’s a reason to slow down, not speed up.
University of North Carolina Press posted its own set of tips recently, noting that scams targeting authors continue to rise. When university presses start writing public safety guidance for their own writers, the volume has crossed a threshold.
What a bot can actually do here
This is where my day job gets useful. You don’t need machine learning to catch most of this. You need a lookup table and some discipline.
Start with an allowlist, not a blocklist
Blocklists lose. New domains are free and infinite. Allowlists win because the set of organizations you actually do business with is small and slow-changing. A simple script that tags any inbound message from outside your known-domains list is a solid first filter. Everything outside the list gets treated as a stranger until proven otherwise.
Automate the lookup, not the decision
I’d build a small bot that takes a sender domain and returns three things: registration age, whether the domain matches a known publisher or agency record, and whether the display name and domain disagree. Then the bot stops. It hands you a report and you make the call. The failure mode of automating the decision itself is that a false negative costs you your rights or your money.
Keep a local record of every claim
Writer’s Digest’s 2026 roundup of websites for writers points to Duotrope, whose paid tier includes a searchable database of current fiction, poetry, nonfiction, and visual art publishers and agents, plus a deadline calendar and a personal submission tracker. That last piece matters more than people realize. A tracker is a source of truth. When someone emails claiming a prior relationship or a pending claim on your behalf, a local record lets you check the assertion instead of trusting your memory under pressure.
The uncomfortable overlap
The thing that makes this moment tricky for authors is that real publishers and real agents are also making claims right now, and authors are contesting some of them. So the naive heuristic of “anyone contacting me about the settlement is a scammer” doesn’t hold. You can’t pattern-match your way out of this on vibes. You need process.
That’s the transferable lesson for anyone building systems in an AI-adjacent space. When the incentives get large enough, your trust boundaries get tested by people who read your documentation more carefully than you did. Authors are learning that the hard way. Build the verification step before you need it, because the message that finally requires it will arrive looking completely ordinary.
đź•’ Published: