Teams already labels the stranger. Unit 42's own figure shows the warning Microsoft raises when an external identity opens a chat with a user [12]. Chats still landed in front of more than 150 employees at 10 or more organizations, an average of at least 15 targets per company [16], which reads as spraying inside each tenant rather than selecting individuals.
The second variant is the part that moves the control surface. Unit 42 says no software exploit is involved: the operators rely on live voice and on the trust employees place in SaaS collaboration platforms [9]. It also says that once the trust gap is crossed, the walk to domain-level privileges using open-source tools such as PetitPotam is short [11]. Put those together and the chain runs on configuration at both ends, namely who may open a chat into the tenant, and whether authentication relayed at a domain controller is accepted. Neither question resolves on the endpoint.
The trend figures deserve a closer read than the percentages suggest. Collaboration-tool phishing went from 30 percent to 42 percent of phishing alerts in Cortex across consecutive four-month periods [6], which is 12 points absolute and a 40 percent relative rise [14]. That is Palo Alto's own telemetry from its own install base, published in a post that also lists Cortex and Idira products as the mitigation [13]. KnowBe4's 41 percent covers October 2025 to March 2026, overlapping Unit 42's January to April window by three months [15], so the two numbers are one directional reading of a shared quarter rather than independent corroboration.
The outcome itself goes unreported. Unit 42 describes attempted payload delivery across two campaigns using two distinct vectors [5], gives no count of employees who executed anything, and names no threat group behind Spring Ring [19]. APT29, cited as prior art, ran Teams abuse the older way, with malicious links and fake Entra ID tenants aimed at harvesting credentials [8]. The delta is real-time audio, which lets the operator pivot on a hesitant victim, and which leaves thinner records than the file operations and mailboxes responders normally review [10]. Unit 42's framing is that identity has become the primary attack vector here [17]; the operational version of that is that the intrusion has a tenant and a federation setting behind it before it has a process tree.
