Silk Typhoon, the group Microsoft first named HAFNIUM, has moved from mass exploitation of on-premises Exchange servers to stealing the API keys and application credentials of IT service providers and using them to walk into downstream customer tenants. That shift turns a supplier's non-human identities into a route into your environment, often without a single phishing email or failed logon. This profile covers the group's names, targeting, the ATT&CK techniques documented against it, how it abuses service principals, OAuth apps and vaulted secrets, and what identity, cloud and SOC teams should monitor and change. It is for defenders running Microsoft 365 and Entra ID, organisations that rely on PAM, remote support or cloud data management vendors, and threat intelligence teams building hunts.
Key takeaways
- Same group, two eras. In 2021 HAFNIUM exploited four Exchange Server zero-days (the ProxyLogon chain); by 2025 Microsoft reported Silk Typhoon abusing stolen API keys and credentials from IT supply chain companies.
- Your supplier's keys are your attack surface. Microsoft describes the group stealing keys from privileged access management, cloud application and cloud data management providers, then reaching those providers' customers.
- Applications are the persistence layer. The group adds its own credentials to already-consented apps, abuses multi-tenant apps and creates apps with names that blend in, then reads mail and files through Microsoft Graph and EWS.
- Edge zero-days and n-days still open the door. Ivanti Connect Secure, PAN-OS GlobalProtect and Citrix NetScaler flaws appear alongside password spraying and passwords leaked on GitHub.
- Detection depends on workload identity telemetry. Service principal sign-in logs, app credential changes and Graph activity are where this actor shows up; many tenants do not collect or review them.
Who they are (names and aliases)
MITRE ATT&CK tracks the group as G0125 HAFNIUM, with associated names HAFNIUM, Operation Exchange Marauder and Silk Typhoon, and describes it as a likely state-sponsored cyber espionage group operating out of China. Microsoft introduced HAFNIUM on 2 March 2021, assessing with high confidence that it was behind the Exchange zero-day campaign, and renamed it Silk Typhoon when it moved to its weather-based naming scheme (Typhoon denotes China-based actors). In a March 2025 post Microsoft said it had tracked the group since 2020.
On 19 July 2021 the United States, with allies, publicly attributed the Exchange compromises to cyber actors affiliated with China's Ministry of State Security, as recorded in CISA advisory AA21-062A. Other vendors use their own names for overlapping activity, and overlaps are imperfect: treat any mapping as approximate. See the Threat Actor Naming Conventions Guide and the China State Cyber Operations Guide for how this group sits alongside Volt, Salt and other Typhoons.
Targeting and motivation
The motive is intelligence collection. Microsoft's 2021 report listed infectious disease researchers, law firms, higher education, defence contractors, policy think tanks and NGOs, primarily in the United States, and noted that the group operated from leased virtual private servers in the US. ATT&CK's description adds a broad set of US sectors.
The 2025 reporting describes a change of method rather than of goal. Microsoft observed Silk Typhoon going after IT providers, remote monitoring and management tools and cloud applications to gain access to many customers at once, and reported victims including state and local government and the IT services sector. For defenders, this means your exposure depends on the security of every vendor holding a key or app registration into your tenant, which is a third-party risk question as much as a SOC one. See the Cybersecurity for MSPs and MSSPs Guide.
Tactics and techniques (MITRE ATT&CK)
The table maps techniques ATT&CK v20 attributes to G0125 (plus one mapping of Microsoft's description, marked) to the procedure reported. It is not exhaustive.
| Tactic | Technique | Documented procedure |
|---|---|---|
| Reconnaissance | T1593.003 Search Open Websites/Domains: Code Repositories | Searched GitHub for leaked corporate passwords |
| Resource development | T1584.005 Compromise Infrastructure: Botnet | Covert networks of compromised devices (Microsoft cites Cyberoam, Zyxel and QNAP) to hide origin |
| Initial access | T1190 Exploit Public-Facing Application | Exchange ProxyLogon chain; Ivanti, PAN-OS and NetScaler vulnerabilities |
| Initial access | T1199 Trusted Relationship | Stolen API keys and credentials of PAM, cloud app and data management providers used against downstream customers |
| Credential access | T1110.003 Password Spraying | Spraying cloud accounts |
| Credential access | T1555.006 Cloud Secrets Management Stores | Read passwords from Azure Key Vault |
| Credential access | T1003.003 OS Credential Dumping: NTDS | Dumped Active Directory |
| Persistence | T1505.003 Web Shell | China Chopper, ASPXSpy and others on Exchange servers |
| Persistence | T1053.005 Scheduled Task, with S1011 Tarrask | Hidden scheduled tasks |
| Persistence | T1098 Account Manipulation; T1136.002 Domain Account | Reset default admin passwords; created domain accounts |
| Persistence (writer mapping) | T1098.001 Additional Cloud Credentials | Microsoft: added credentials to already-consented applications |
| Lateral movement, access | T1078.004 Cloud Accounts; T1550.001 Application Access Token | Abused service principals and app tokens to reach data |
| Collection | T1114.002 Remote Email Collection; T1530 Data from Cloud Storage | Exported mailboxes through Microsoft Graph; took OneDrive and SharePoint content |
| Exfiltration | T1567.002 Exfiltration to Cloud Storage | Uploaded data to MEGA |
| Defence impairment | T1685.005 Clear Windows Event Logs | Cleared logs after activity |
Vulnerabilities associated with the group
| CVE | Product | Nature | Added to CISA KEV |
|---|---|---|---|
| CVE-2021-26855 | Exchange Server | Server-side request forgery, first link of ProxyLogon | 3 November 2021 |
| CVE-2021-26857 | Exchange Server | Insecure deserialisation in the Unified Messaging service | 3 November 2021 |
| CVE-2021-26858, CVE-2021-27065 | Exchange Server | Post-authentication arbitrary file write | 3 November 2021 |
| CVE-2023-3519 | Citrix NetScaler ADC and Gateway | Code injection | 19 July 2023 |
| CVE-2024-3400 | Palo Alto Networks PAN-OS | Command injection | 12 April 2024 |
| CVE-2025-0282 | Ivanti Connect Secure, Policy Secure, ZTA Gateways | Stack-based buffer overflow | 8 January 2025 |
Microsoft listed the 2023 to 2025 CVEs among those the group exploited; the KEV dates show when CISA confirmed exploitation in the wild by anyone, not when Silk Typhoon first used them.
How they abuse identities and credentials
This is where the profile matters most for identity teams. Microsoft's 2025 reporting describes a chain that rarely touches a human password once the group is inside.
- Supplier key theft: the group compromised IT providers and took API keys and credentials that those providers use to manage their customers. A key built to reset passwords or pull data across many tenants becomes a multi-victim initial access token. See the API Key Management Guide.
- Vault harvesting: ATT&CK records reads from Azure Key Vault, and Microsoft reports the group stealing passwords held in vaulting products. A vault is only as safe as the identities allowed to read it.
- Hybrid identity pivots: Microsoft reports targeting of Microsoft Entra Connect (formerly AAD Connect) servers, which hold the synchronisation account bridging on-premises AD and the cloud. See the Microsoft Entra Connect and Cloud Sync Hardening Guide.
- Application credential backdoors: adding a client secret or certificate to an app that already holds admin consent gives silent, MFA-free access with the app's permissions. Creating new apps named to look like legitimate tooling, and abusing multi-tenant apps, extends the same idea. See the Entra ID Applications and Consent Guide.
- App-only data access: with application permissions such as Mail.Read or Files.Read.All, Graph and EWS let an attacker read every mailbox or drive in scope. Activity appears as service principal sign-ins, not user sign-ins.
- Leaked and sprayed passwords: GitHub searches and spraying supply the first foothold where no vulnerability is available. See the Password Spraying and Credential Stuffing Defence Guide.
The pattern overlaps with the key theft covered in the Storm-0558 and Cloud Key Theft Profile, but where Storm-0558 forged tokens with a stolen signing key, Silk Typhoon mostly uses keys and app credentials that were legitimately issued and then stolen. Rotation, scoping and ownership of those credentials are the controls that matter.
Notable incidents
- Exchange ProxyLogon campaign (2021): Microsoft disclosed on 2 March 2021 that HAFNIUM was exploiting four zero-days in on-premises Exchange to steal mailbox data and plant web shells. CISA's AA21-062A told organisations to investigate from 1 January 2021 and, where exploitation was found, to assume the network identity had been compromised. Exploitation by other actors followed quickly once the flaws were public.
- IT supply chain campaign (reported March 2025): Microsoft described stolen keys from PAM, cloud app and cloud data management companies used to access downstream customer environments, followed by app abuse and Graph exfiltration.
- The PAM vendor key pattern in practice: the BeyondTrust breach of December 2024 shows exactly this risk: a stolen Remote Support SaaS API key able to reset local application passwords was used against customers including the US Treasury, which attributed the intrusion to a China state-sponsored actor. Treasury's statement did not name a specific group, so treat it as an illustration of the technique, not an attribution.
Detection and defence
What to monitor
| Signal | Where to find it | Why it matters |
|---|---|---|
| New client secret or certificate on an existing app or service principal | Entra ID audit log ("Update application, Certificates and secrets management", "Add service principal credentials") | Primary persistence method described by Microsoft |
| Service principal sign-ins from new IPs, regions or hosting ranges | Entra ID service principal sign-in logs | App-only access does not appear in user sign-in logs |
| New app registrations or consents to high-privilege Graph permissions | Entra ID audit log, consent events | Apps named to blend in with legitimate tooling |
| Bulk mailbox or file reads by an app | Microsoft Graph activity logs, unified audit log (MailItemsAccessed) | Collection through Graph and EWS |
| Reads from key vaults by unusual identities | Azure Key Vault diagnostic logs | T1555.006 vault harvesting |
| Exchange exploitation artefacts | HttpProxy logs with empty AuthenticatedUser and AnchorMailbox like ServerInfo~*/*, ECP logs showing Set-*VirtualDirectory with script, new .aspx files in aspnet_client or owa\auth | Indicators Microsoft published for ProxyLogon |
| Security event log clearing | Windows Event ID 1102, System Event ID 104 | T1685.005 |
Hardening priorities
- Inventory every external key and app: list all vendor API keys, service principals, multi-tenant apps and delegated admin relationships with access to your tenant, with an owner and a business purpose for each. See the GDAP and MSP Delegated Admin Access Guide.
- Cut app-only permissions down: Microsoft recommends reviewing application permissions and using RBAC for Applications in Exchange Online to scope mail access to specific mailboxes rather than the whole tenant.
- Apply Conditional Access to workload identities: restrict where single-tenant service principals can sign in from and alert on risky service principal sign-ins.
- Harden hybrid identity: treat Entra Connect servers as Tier 0 and avoid on-premises accounts holding cloud admin roles, as Microsoft advises.
- Patch internet-facing appliances against KEV: where a listed CVE was exposed before patching, hunt for compromise; patching alone does not remove a web shell or a stolen key.
- Rotate and scope vendor keys: require IP allow-listing, expiry and least privilege on keys you issue to suppliers, and contract for prompt notification of key compromise.
Response
If you find a malicious credential on an app, remove it, rotate every secret on that app, review all permissions it holds and search Graph and audit logs for its activity back to the start of retention. If a supplier reports key theft, revoke the key, reset anything the key could reset, and review access from that supplier's identities. See the OAuth Token Theft Incident Playbook, the Web Shell Response Playbook and the Supply Chain Compromise Response Playbook.
Metrics
- Owned workload identities: percentage of service principals and vendor keys with a named owner and review date.
- Credential age: number of app secrets older than your rotation policy, and number of apps holding more than one active secret.
- High-privilege app permissions: count of apps with tenant-wide Mail, Files or Directory write permissions, tracked downward.
- Telemetry coverage: service principal sign-in and Graph activity logs collected and retained for at least the period your investigations need.
- KEV exposure time: days from KEV listing to patch for internet-facing appliances.
Framework mapping
| Defensive need | NIST SP 800-53 Rev. 5 | NIST CSF 2.0 |
|---|---|---|
| Inventory and lifecycle of workload identities and keys | AC-2 Account Management, IA-5 Authenticator Management | PR.AA-01 |
| Least privilege for app permissions | AC-6 Least Privilege | PR.AA-05 |
| Supplier access and key handling | SA-9 External System Services, SR-3 Supply Chain Controls and Processes | GV.SC-07 |
| Edge vulnerability management | RA-5 Vulnerability Monitoring and Scanning | ID.RA-01 |
| Monitoring of sign-ins and app activity | SI-4 System Monitoring, AU-6 Audit Record Review | DE.CM-01 |
Practitioner checklist
- Export every service principal and app registration with credentials, permissions, owner and last sign-in; disable those without an owner.
- Alert on any new secret or certificate added to an existing application, and on new consents to Graph application permissions.
- Collect service principal sign-in logs and Microsoft Graph activity logs into your SIEM.
- Scope Exchange Online app access with RBAC for Applications instead of tenant-wide Mail permissions.
- List every vendor that holds an API key or delegated admin access to you, and confirm rotation, IP restriction and breach notification terms.
- Treat Entra Connect servers and key vaults as Tier 0 and restrict who can read vault secrets.
- Patch the CVEs in the table above on any internet-facing system and hunt where exposure preceded patching.
- Scan your public code repositories for leaked credentials and rotate anything found.
- Run a tabletop on "a supplier tells us its API key into our tenant was stolen".
Standards and references
- MITRE ATT&CK: HAFNIUM (G0125)
- Microsoft Threat Intelligence: Silk Typhoon targeting IT supply chain
- Microsoft Threat Intelligence: HAFNIUM targeting Exchange Servers with 0-day exploits
- CISA: AA21-062A Mitigate Microsoft Exchange Server Vulnerabilities
- CISA: Known Exploited Vulnerabilities Catalog
- NIST: SP 800-53 Rev. 5
- NIST: Cybersecurity Framework 2.0
Related NHI Mgmt Group resources: Storm-0558 and Cloud Key Theft Profile · Azure RBAC and Service Principals Guide · SaaS and OAuth App Governance Guide · Volt Typhoon Threat Actor Profile · Compromised MSP Playbook