Attackers who broke into corporate networks through Microsoft Teams did it by pretending to be the IT help desk. Microsoft's report on the incidents describes Teams messages and calls from people posing as support staff, convincing employees to open Quick Assist and hand over the connection code, which gave the intruders keyboard control of the workstation. CNET covered the report and Microsoft's response.
From there the work turned ordinary. The attackers installed tooling on Python, Bun and Java runtimes, opened network paths and granted themselves authentication, which is enough to move from one laptop into the rest of the environment. Microsoft's writeup says both incidents began with the same Teams phishing message and call, and that the social engineering, not a software flaw, produced the access.
The vulnerability was a phone call
Nothing in the chain required a zero-day. Quick Assist is a legitimate Windows tool that exists so a help desk can fix a machine without driving to it. The attackers used it exactly as designed, with a user who believed the person on the other end was doing their job.
Yu Cai, associate director of the Institute of Computing and Cybersystems at Michigan Tech, told CNET the simplicity was the surprising part. Sophisticated intrusions do not have to start with sophisticated exploits, he said, and this one began with employees trusting a message that arrived through a tool they use every day.
That framing matters because the industry has spent two years pointing at AI as the source of the next wave of attacks. The same week, the more effective intrusion was a person reading a script. Cai's point is that the two are not separate problems: voice cloning and generated video make the same call more convincing, so the human attack gets cheaper to run rather than rarer.
This is the same pattern as other recent breaches that skipped the exploit entirely. A remote access trojan talked Android into handing over admin rights, and North Korea hid malware inside coding tests sent to 30,000 machines. In each case the technical control worked; the person in front of it was the way in.
Detection is arriving after the conversation starts
Microsoft says it is adding phishing detection that reads conversation context to spot social engineering in higher-risk messages, particularly first contact from external users. Its roadmap also lists a way to report suspicious guest invitations and a change that blurs QR codes sent from outside the organisation. Administrators can already block external Teams users or restrict who is allowed to chat with people outside the company.
None of that is a switch that closes the hole. Context-based detection only has something to read once the attacker has already opened the conversation, which is the moment the report says the damage begins. Blurring QR codes removes one delivery trick and leaves the phone call intact.
The control that would have stopped both incidents is the one that is hardest to enforce: a rule that help desk staff never cold-call a user and ask them to open a remote session, and that any such request gets verified through a second channel. That is a policy, not a product, and it has to survive the busy Tuesday when an employee is behind on tickets and a friendly voice offers to fix the laptop.
Microsoft's statement that users should verify a sender's identity before acting pushes the decision down to the person least equipped to make it. The company has not said whether it will stop Quick Assist sessions initiated from an inbound call, which is the specific mechanism both intrusions used. Until something changes there, the intrusion path stays open, and the only question is how many employees recognise the script before they read out the code.