Two case studies in ASD’s Annual Cyber Threat Report 2025–26 describe the same attack from opposite ends. One is a joint advisory on Scattered Spider. The other is an Australian bank’s account of two phone calls to its internal help desk. Neither involves an exploit, a malicious attachment or a zero-day. Both go through the one function whose job is to make access easier.
How the call works
ASD’s description of Scattered Spider’s method is worth reading slowly. The group researches a target organisation for usernames, personal details and phone numbers, sometimes performing SIM swaps along the way. Then it calls. The first calls are not attacks. They are reconnaissance, designed to learn what the help desk needs to hear before it will reset a password. Later calls gather the specific details for a chosen employee. The final call asks the help desk to reset that employee’s password or transfer their multi-factor authentication to a new device. After that, the attacker logs in through single sign-on as the employee, with every system trusting them.
The Australian bank’s case shows the reconnaissance phase from the inside. In August 2025 two callers rang its help desk impersonating a Melbourne-based employee. They had the employee’s staff ID number. They said they could not reach colleagues on Microsoft Teams. They were confrontational and pushed hard. They left with the names, job titles and direct phone numbers of several staff in a specialised financial services team. The bank traced the calls to a call-centre environment in Hong Kong.
Two details in the bank’s account deserve attention. First, the targeted team was not public-facing, and the bank found that made help desk staff less suspicious, not more. Nobody expects an outsider to know who works in a back-office team. Second, the bank reported the incident to ASD because of recent media coverage of help desk vishing. Without that coverage, an attack that harvested a target list from inside a bank might have been logged as a difficult caller.
The numbers behind it
ASD’s mapping of incidents to MITRE ATT&CK puts phishing first, at 39% of incidents. Behind it come compromised accounts at 21%, user execution at 17% and valid accounts at 15%. The report’s own summary is blunt: attackers steal access, not just data, and compromised accounts and credentials are now among the most common incident categories across government, industry and critical infrastructure.
One figure looks like good news and probably is not. Incidents involving “Gather Victim Identity Information” fell from 30% to 13% year on year. ASD reads this as a shift from reconnaissance towards direct access. A more uncomfortable reading is that reconnaissance has moved somewhere incident data cannot see it. A phone call to the help desk that ends with a polite refusal, or a polite answer, is rarely recorded as an incident at all. The bank’s case is in the report precisely because it was the exception.
MFA is only as strong as its reset path
Every board has been told to enforce multi-factor authentication, and rightly. ASD repeats the advice in this report. But MFA protects the login, not the process that issues and replaces the second factor. If a convincing caller can have a token moved to a new phone, the organisation’s MFA is ultimately protected by whatever the help desk asks before it agrees.
Where that check relies on facts about the employee, such as staff ID, date of birth, manager’s name or cost centre, every one of those facts is knowable. Staff IDs appear on email signatures, badges and leaked datasets. Org charts can be rebuilt from LinkedIn. AI has made that rebuild faster and the resulting phone script more convincing, as ASD notes throughout the report.
If the help desk is measured only on time to resolve, first-call resolution and user satisfaction, it is rewarded for saying yes quickly. Under those measures, a service desk agent who refuses an angry caller claiming to be locked out before a deadline is, by their own KPIs, underperforming. Those incentives favour the attacker, while the agent is left as the last line of defence.
ASD’s own process shows the alternative. If someone receives a call claiming to be from ASD’s ACSC, the report tells them to hang up, ring 1300 CYBER1 and quote the reference number. The person who receives the call verifies through a channel the caller does not control. A help desk that verifies a caller using details the caller supplies does the reverse.
What boards should ask for
Treat credential reset and MFA re-enrolment as privileged operations. They grant the same access as a stolen password and token combined. They should carry the controls you would apply to any privileged action: a named control owner, defined approval steps, and logging that someone reviews.
Verify through a channel the caller does not control. Call back on the number held in the HR system, not the number offered. Require approval from the user’s manager through a separate channel. For privileged and executive accounts, require in-person or video verification against identity documents, and accept the delay.
Alert on resets, not just logins. An MFA device change followed shortly by a login from a new location is one of the clearest compromise signals available. Identity platforms typically record both events. The signal only appears when someone correlates them.
Fix the help desk incentives. If agents are measured on speed and satisfaction alone, change the scorecard so that a correctly refused request counts as a good outcome. Agents need explicit permission, in writing, to say no to someone senior.
Bring outsourced service desks into supply chain risk. Where a managed service provider runs the help desk, its verification procedure is your access control. For APRA-regulated entities, that provider’s practices sit squarely within CPS 230’s expectations for material service providers.
Test it. Include help desk vishing in red team scope, with the board’s knowledge, and report the result. If nobody has ever tried to talk your service desk into a password reset, you do not know whether they would.
Report the near misses. The bank’s case reached ASD by luck of timing. Make it policy that suspicious help desk calls are reported to security and, where they show targeting, to ASD’s ACSC.
The perimeter moved
For most of the past decade, security investment followed the network edge: firewalls, email filtering, endpoint protection. Those controls are now good enough that the cheapest path in is to ask a person. The attackers in ASD’s case studies did not defeat a single technical control. They found the process that existed to bypass those controls for legitimate users, and they became a legitimate user. The question for boards is whether anyone in the organisation owns that process as a security control, or whether it is still owned by whoever owns the service desk.