Who they are and who they target
APT33 is a long-running, Iran-linked espionage group known by several names: Elfin at Symantec (now Broadcom), Peach Sandstorm at Microsoft (formerly Holmium), Refined Kitten at CrowdStrike, and Curious Serpens at Palo Alto Networks Unit 42. FireEye named APT33 in September 2017 and described a suspected Iranian group active since at least 2013. MITRE ATT&CK records the same start date and the same core interest: aviation and energy, with victims in the United States, Saudi Arabia and South Korea. Our earlier APT33 (Elfin) analysis covers the 2017-era campaign history in more depth.
Assessment date: 21 November 2026. This profile reflects reporting published up to that date; the 2023 and 2024 campaigns are historical baseline, not evidence of current operations.
Targeting has widened without changing character. FireEye focused on aerospace and energy; Microsoft’s later reporting adds satellite operators, communications equipment, the Defense Industrial Base and government, plus higher education — used less as a target than as a source of stolen Azure subscriptions. Symantec’s 2019 update described Elfin spreading across government, research, chemical, engineering, manufacturing and telecoms in the US and Saudi Arabia.
What is confirmed: the intrusions, the tooling, the victimology. What is assessed: who is behind them. FireEye called the group “suspected Iranian” and found ties to the Nasr Institute but no link to a specific government agency at the time (Reuters, 2017). Microsoft goes further, assessing that Peach Sandstorm operates on behalf of the Islamic Revolutionary Guard Corps “based on the group’s victimology and operational focus” — an analytic judgment from targeting patterns, not documentary proof (Microsoft Security Blog, 28 August 2024). Alias merging is also judgment: vendors cluster activity by shared tooling and infrastructure, and those links are probabilistic. We know of no public indictment or sanction naming APT33. Treat the Iran/IRGC line as well supported but assessed, and say so in briefings.
FireEye also assessed in 2017 that APT33 had potential destructive capability and possible ties to the Shamoon and StoneDrill wipers — low confidence, never demonstrated. It belongs in planning, not daily detection.
Known tooling and tradecraft
The throughline is credentials. APT33 rarely needs a zero-day; it needs one weak password and a reachable login portal.
Initial access. Password spraying is the signature. Microsoft observed campaigns against thousands of organizations from February 2023 onward, run from Tor exit nodes, carrying a distinctive go-http-client user agent, with activity bunched into Iranian business hours. Alongside it sit the older techniques: spearphishing on job and career lures that led to .hta payloads (FireEye, 2017), and fake LinkedIn personas posing as students, developers and recruiters (Microsoft, 2024). Exploitation is opportunistic, aimed at internet-facing appliances with public exploits — Confluence CVE-2022-26134 and ManageEngine CVE-2022-47966 in the 2023 wave, WinRAR CVE-2018-20250 in Symantec’s 2019 reporting.
Cloud tradecraft. Once inside an identity, the group behaves like a cloud administrator. Microsoft documented AzureHound and ROADtools reconnaissance, new Azure subscriptions inside victim tenants, and persistence through Azure Arc — enrolling an on-premises device into an attacker-controlled subscription. A March 2023 intrusion used a Golden SAML attack: stolen AD FS token-signing keys minting trusted SAML tokens to bypass federation. Other campaigns created Azure tenants and student subscriptions to host command-and-control.
Malware and tooling. FireEye-era tooling included TURNEDUP, Nanocore and NetWire. Later waves brought custom backdoors: FalseFont against defense contractors in December 2023, and Tickler between April and July 2024 — a multi-stage backdoor shipped as a .pdf.exe inside a ZIP with decoy PDFs, using DLL sideloading of a signed Microsoft.SharePoint.NativeMessaging.exe, a Run key for persistence, and command-and-control on azurewebsites.net app services. Symantec documented DarkComet, LaZagne, Mimikatz and SniffPass earlier. The pattern holds: legitimate binaries, legitimate cloud services, custom code only where it buys something.
Techniques and detection
Each item gives the ATT&CK technique, the telemetry that matters, and what to look for. Run them as scheduled hunts.
- T1110.003 — Password Spraying. Telemetry: Entra ID, Okta or VPN sign-in logs. Look for one or two passwords tried across many accounts, low failure counts per account, a single ISP or ASN distributing the attempts, Tor egress, the go-http-client user agent, and attempts compressed into Iran business hours. Sentinel baseline:
SigninLogs | where TimeGenerated > ago(7d) | where ResultType == "50126" | summarize Accounts = dcount(UserPrincipalName), Attempts = count(), IPs = dcount(IPAddress) by bin(TimeGenerated, 1h), AppDisplayName, UserAgent | where Accounts >= 25 and Attempts >= 100 - T1078.004 — Valid Accounts: Cloud Accounts. A spray that succeeds becomes a cloud intrusion. Alert on successful sign-ins that follow a spray wave from commercial VPN ranges, new subscription or tenant creation, Azure Arc onboarding, OAuth consent grants, MFA registration and inbox forwarding rules. On any hit, reset the password, revoke sessions and refresh tokens, then re-challenge MFA.
- T1566.002 and T1204.001 — Spearphishing Link and User Execution. Career and HR themes remain the lure of choice, now often delivered through LinkedIn or Teams rather than email. Hunt for double extensions (.pdf.exe, .pdf.lnk), archives containing executables, .hta links, and external Teams file transfers to engineering and program staff. Block double extensions and HTA execution at the gateway and endpoint.
- T1574.002 — DLL Side-Loading. Tickler abused a signed Microsoft.SharePoint.NativeMessaging.exe to load unsigned DLLs. On Sysmon Event ID 7, flag signed binaries running outside Program Files that load unsigned modules from Downloads, Temp or user profile paths.
- T1547.001 — Registry Run Keys. The same campaign’s batch script wrote a Run key pointing at SharePoint.exe. Alert on Run key creation (Sysmon Event ID 13) where the value resolves to a user-writable directory and the parent process is cmd, wscript or a script host.
- T1606.002 — Forge Web Credentials: SAML Tokens. The Golden SAML path is quiet by design. Monitor AD FS token-signing certificate export, replacement or DKM key access, and alert on AD FS changes outside a change window. If keys were exposed, rotate them and invalidate refresh tokens.
- T1071.001 — Web Protocols. Tickler’s command-and-control sat on azurewebsites.net app services with names built from support and satellite vocabulary. Alert on rare FQDNs in that namespace, and on HTTP POST beaconing from hosts with no business calling Azure app services:
DeviceNetworkEvents | where Timestamp > ago(30d) | where RemoteUrl endswith ".azurewebsites.net" | summarize Hosts = dcount(DeviceId), FirstSeen = min(Timestamp), LastSeen = max(Timestamp) by RemoteUrl | where Hosts <= 2
Mitigation priorities
- Make the password spray unprofitable. Enforce MFA everywhere, block legacy authentication, enable smart lockout and banned-password lists, and alert on spray patterns. Passwordless sign-in for privileged accounts removes the technique entirely.
- Treat AD FS as Tier 0. Restrict administrative access, log certificate operations, and treat token-signing key exposure as a full identity compromise. Where possible, migrate federation to Entra ID authentication.
- Put guardrails on cloud self-service. Limit who can create subscriptions and tenants, alert on Azure Arc enrollment and OAuth consent, and review app registrations monthly. These moves stop a stolen account from becoming a foothold.
- Control execution. Application control or WDAC on servers, plus attack surface reduction rules blocking executables from email and downloads folders, kills most of the payload chain early.
- Watch egress, not just ingress. Proxy and firewall logs should surface first-seen azurewebsites.net and similar platform-as-a-service domains, and any RMM tool not on an approved list.
- Patch the edge and rehearse the wiper. Inventory internet-facing Confluence, ManageEngine and file-transfer appliances, and patch on a schedule you can prove. Keep offline backups and a wiper-specific tabletop; the destructive capability is unproven, but the cost of being wrong is total.
What to watch for next
Regional tension is the forcing function. In 2026 the UAE’s cyber leadership described Iran-linked attacks on electricity, water and desalination plants using wiper malware, and said national teams had contained organized attacks on aviation, energy and education in August (The National, 18 November 2026). That victim profile overlaps APT33’s, but no public source has attributed those incidents to APT33. Treat the overlap as a reason to check exposure, not as an identification.
The direction of travel is visible in the reporting. Unit 42’s 2026 incident response report notes that an Iran-linked cluster it tracks as Curious Serpens — mapped to APT33 by CERT-EU — kept using employment-themed lures against aerospace and satellite-communications providers through 2025. Combined with the 2024 shift to custom backdoors and cloud-hosted command-and-control, expect fewer commodity tools, more identity abuse, and delivery containers built to dodge Mark-of-the-Web.
- New azurewebsites.net or app-service domains built from support, satellite or helpdesk vocabulary.
- Spray waves from Tor or commercial VPN ranges using go-http-client, especially against education tenants.
- Subscription creation, Azure Arc enrollment or Golden SAML indicators in tenants with no cloud engineering team.
- Recruitment-themed approaches on LinkedIn or Teams that move to a file transfer within days.
- Archive and disk-image attachments containing executables, including .iso and .vhd containers.
Two limits. First, no APT33-specific reporting appeared in the current news cycle, so this is a baseline, not a live feed. Second, the Iran/IRGC attribution is a vendor assessment built on victimology and tradecraft; it is strong enough to act on and not strong enough to state as fact. Review this profile quarterly, re-run the hunts monthly, and check attribution claims against primary reporting before they reach a board slide.
Sources:
- MITRE ATT&CK, APT33 (G0064): attack.mitre.org/groups/G0064
- FireEye, “Insights into Iranian Cyber Espionage,” September 2017: fireeye.com/blog/threat-research/2017/09/apt33-insights-into-iranian-cyber-espionage.html
- Microsoft, Peach Sandstorm reporting, 2023–2024: microsoft.com/en-us/security/blog
- Symantec, Elfin/APT33 analysis, 2019: symantec.com/blogs/threat-intelligence/elfin-apt33-espionage
- CERT-EU Cyber Security Brief 24-04, March 2024: cert.europa.eu/publications/threat-intelligence/cb24-04
- The National, UAE cyber chief on Iran-linked attacks, 18 November 2026: thenationalnews.com
- Unit 42, 2026 Incident Response Report: paloaltonetworks.com/resources/research/unit-42-incident-response-report