Charming Kitten, commonly tracked as APT35 and overlapping in public reporting with APT42, Mint Sandstorm, Magic Hound, and TA453, has repeatedly targeted researchers, journalists, diplomats, policy professionals, activists, and others whose accounts provide access to sensitive information or influential networks.

The most useful way to understand this activity is through the sequence that links target selection, relationship building, authentication, and collection.

Our analysis of public reporting from 2018 to 2026 shows that professional trust plays a central operational role. The actor repeatedly uses credible personas, conference invitations, media requests, research exchanges, and other context-specific approaches to establish legitimacy. Once that relationship feels familiar, the target may be asked to open a document, move to another communication channel, approve a login, or authenticate to a service that appears legitimate.

This makes the attack especially effective against knowledge-producing communities. Their work depends on openness, rapid collaboration, and contact with unfamiliar people. The adversary can exploit those same habits as a route to access.

A second important finding concerns target selection. The recurring victims do not belong to one sector. What connects them is their position inside information networks. A mid-level researcher with access to draft policy work, a journalist with a valuable source network, or an event organizer with private participant lists may be more useful than a senior executive with heavily protected systems.

The report also shows why personal and enterprise identities cannot be treated as separate worlds. A professional conversation may begin in a corporate mailbox and continue through a personal account or messaging app. The same persona and pretext can move across channels while preserving trust, creating gaps in visibility for defenders.

The March 2026 think-tank campaign illustrates the model clearly. The approach began with a legitimate-looking invitation and a benign document hosted on OneDrive. The credential-capture stage appeared later, after the relationship had already been established.

For defenders, this changes the problem. Blocking domains and training users to spot poor grammar remain useful, but they address only the final stages of the operation. More resilient defenses start earlier and continue after authentication. High-risk users need phishing-resistant authentication, fast ways to verify unfamiliar professional contacts, monitoring for mailbox and cloud changes, and incident-response procedures that treat identity compromise as a complete espionage event even when no malware is deployed.

The broader lesson is that modern espionage can succeed through ordinary professional behavior. The critical security boundary is often the point where a credible human interaction becomes a request for digital trust.