ESET takes half in international operation to disrupt Lumma Stealer

ESET has collaborated with Microsoft, BitSight, Lumen, Cloudflare, CleanDNS, and GMO Registry in a worldwide disruption operation in opposition to Lumma Stealer, an notorious malware-as-a-service (MaaS) infostealer. The operation focused Lumma Stealer infrastructure with all identified C&C servers previously yr, rendering the exfiltration community, or a big a part of it, nonoperational.

Key factors of this blogpost:

• ESET took half in a coordinated international operation to disrupt Lumma Stealer.
• ESET supplied technical evaluation and statistical info, and extracted important knowledge from tens of hundreds of malware samples.
• We offer an outline of the Lumma Stealer MaaS ecosystem.
• We additionally present technical evaluation and an outline of the evolution of Lumma Stealer’s key static and dynamic properties, which had been vital to the disruption effort.

Disruption contribution

ESET automated techniques processed tens of hundreds of Lumma Stealer samples, dissecting them to extract key parts, comparable to C&C servers and affiliate identifiers. This allowed us to constantly monitor Lumma Stealer’s exercise, monitor growth updates, cluster associates, and extra.

Infostealer malware households, like Lumma Stealer, are sometimes only a foreshadowing of a future, far more devastating assault. Harvested credentials are a valued commodity within the cybercrime underground, offered by preliminary entry brokers to numerous different cybercriminals, together with ransomware associates. Lumma Stealer has been one of the prevalent infostealers over the previous two years, and ESET telemetry (see Determine 1) confirms that it has left no a part of the world untouched.

Lumma Stealer builders had been actively creating and sustaining their malware. Now we have commonly seen code updates starting from minor bug fixes to finish substitute of string encryption algorithms and modifications to the community protocol. The operators additionally actively maintained the shared exfiltration community infrastructure. Between June 17^th, 2024 and Might 1^st, 2025, we noticed a complete of three,353 distinctive C&C domains, averaging roughly 74 new domains rising every week together with occasional updates to Telegram-based dead-drop resolvers (see Determine 2). We talk about the small print of the community infrastructure later within the blogpost.

This ongoing evolution underscores the numerous menace posed by Lumma Stealer and highlights the significance and complexity of the disruption effort.

Background

Over the previous two years, Lumma Stealer (also referred to as LummaC or LummaC2) has emerged as one of the lively infostealers within the cybercrime ecosystem, changing into a popular instrument amongst cybercriminals as a consequence of its lively growth of malware options and its infrastructure being offered as a service.

Malware as a service

Lumma Stealer adopts the idea of malware as a service (MaaS), the place associates pay a month-to-month price, based mostly on their tier, to obtain the newest malware builds and the community infrastructure obligatory for knowledge exfiltration. Associates have entry to a administration panel with a user-friendly interface the place they’ll obtain exfiltrated knowledge and harvested credentials.

The tiered subscription mannequin ranges from USD 250 to USD 1,000 per 30 days, every with more and more refined options. Decrease tiers embody fundamental filtering and log obtain choices, whereas greater tiers supply customized knowledge assortment, evasion instruments, and early entry to new options. The costliest plan emphasizes stealth and adaptableness, providing distinctive construct era and decreased detection.

The operators of Lumma Stealer have additionally created a Telegram market with a ranking system for associates to promote stolen knowledge with out intermediaries. {The marketplace} has been properly documented in Cybereason analysis. Furthermore, they keep public documentation of the administration panel for associates and periodically share updates and fixes on hacking boards, as proven in Determine 3.

Open documentation not solely helps associates with much less expertise to make use of the malware service, but in addition gives invaluable insights for safety researchers. Builders give attention to malware builds, knowledge pipelining, and infrastructure upkeep, whereas associates are answerable for distributing the malware. This info, mixed with the service’s reputation, ends in all kinds of compromise vectors.

Frequent distribution strategies embody phishing, cracked software program, and different malware downloaders together with SmokeLoader, DarkGate, Amadey, Vidar, and others. Fashionable phishing schemes contain ClickFix or pretend CAPTCHA net pages, fraudulent boards with cracked software program, pretend GitHub repositories, fraudulent hyperlinks on Reddit boards, and lots of extra.

Technical evaluation

Quite a few public analyses have already been written about Lumma Stealer and its compromise vectors. Our focus right here, nevertheless, is on the elements related to the disruption. On this part, we’ll briefly introduce the important thing static and dynamic properties that we’ve been actively extracting from Lumma Stealer.

Static properties of Lumma Stealer

Numerous info comes embedded in Lumma Stealer malware samples. This naturally presents a super goal for automated extraction. Apart from the apparent knowledge of curiosity – C&C server domains – the samples additionally include identifier strings that tie the pattern to a particular affiliate and a marketing campaign, and an optionally available identifier resulting in a customized dynamic configuration. These identifiers are utilized in community communication with the C&C server throughout knowledge exfiltration and requests for dynamic configuration. Within the sections under, we take a look at these properties in depth.

C&C domains

Every Lumma Stealer pattern accommodates an inventory of 9 encrypted C&C domains. Whereas the encryption strategies have advanced over time, the attribute array construction has remained constant as much as the time of writing.

Primarily based on Lumma Stealer’s inside pattern versioning, which is closely protected by stack string obfuscation, we all know that up till January 2025, the C&C domains within the samples had been protected by an XOR operate and base64 encoding (see Determine 4). When the base64-encoded string was decoded, it revealed a construction the place the primary 32 bytes served as an XOR key, and the remaining bytes contained the encrypted C&C area.

In January 2025, Lumma Stealer transitioned the safety of the C&C listing to ChaCha20 encryption with a single hardcoded key and nonce (see Determine 5). This safety of the C&C listing within the Lumma Stealer binaries has remained the identical up till the time of publication.

Useless-drop resolvers

Since June 2024, every Lumma Stealer construct got here with a brand new characteristic for acquiring a backup C&C. If no C&C server from the static config responds to Lumma Stealer, then the backup C&C is extracted from a dummy Steam profile net web page appearing as a dead-drop resolver. The Steam profile URL is closely protected within the binary, the identical method because the model string. The encrypted backup C&C URL is ready within the Steam profile identify, as proven in Determine 6, and the safety is an easy Caesar cipher (ROT11).

In February 2025, Lumma Stealer acquired an replace that included a characteristic for acquiring a brand new, major C&C URL from a Telegram channel dead-drop resolver. The C&C URL is extracted from the Telegram channel’s title subject, and it’s protected by the identical algorithm as within the case of the Steam profile dead-drop resolver. The primary distinction within the utilization of the Telegram and Steam profile dead-drop resolvers is that the Telegram choice is examined first, whereas the Steam profile is used as a final resort if profitable communication has not been established with beforehand obtained C&C servers (Determine 16).

Furthermore, we imagine that the Telegram dead-drop resolver is accessible for greater tier subscriptions. It’s because many samples would not have the Telegram URL set, and due to this fact the malware skips this technique.

Lumma Stealer identifier

Every Lumma Stealer pattern accommodates a singular hardcoded affiliate identifier generally known as LID. It’s embedded in plaintext type and utilized for communication with C&C servers. Up till March 2025, the LID parameter string adopted a structured format, delimited by two dashes (Determine 7). A detailed evaluation of the LID affiliate string is supplied in an upcoming part.

Though probably the most prevalent LID noticed throughout our monitoring begins with the string uz4s1o; the second commonest LID, which begins with LPnhqo, gives a greater instance for visualizing typical LID variability. Within the phrase cloud in Determine 8, we current the highest 200 LIDs collected throughout our monitoring, beginning with LPnhqo.

Nevertheless, in early March 2025, Lumma Stealer transitioned to utilizing hexadecimal identifiers, referred to internally as UID (see Determine 9).

Non-obligatory configuration identifier

Along with the LID parameter, Lumma Stealer samples may include an optionally available parameter referred to internally as J. When current, this parameter is in cleartext and formatted as a 32-byte ASCII hex string (see Determine 10). The J parameter is utilized within the C&C request for dynamic configuration with extra definitions for exfiltration. We speak about dynamic configuration in additional element in a following part.

If the J parameter is lacking within the Lumma Stealer pattern, an empty string is used within the C&C request and a default configuration is retrieved. Not like LID, the J parameter isn’t current in Lumma Stealer samples. Nevertheless, it performs a vital position when current, because it permits retrieving a dynamic configuration that considerably will increase the stealer’s capabilities, making it a extra versatile exfiltration instrument for menace actors.

In March 2025, when the LID parameter was renamed to UID and its format modified, the J parameter was renamed to CID however with no change to its format or operate.

Evaluation of static properties

From our long-term monitoring and statistical evaluation of LID parameters, we imagine that the primary phase of the LID identifies the affiliate, whereas the second phase differentiates between campaigns. Primarily based on this assumption you possibly can see the highest 200 affiliate identifiers in Determine 11.

Furthermore, we’ve been in a position to create a visualization of the associates’ actions over the previous yr (see Determine 12). This visualization highlights per week in January 2025. These kinds of visualizations have supplied us with invaluable insights into the patterns and behaviors of various menace actors. Moreover, the visualizations reveal a shared, domain-based C&C infrastructure amongst most Lumma Stealer associates. On the identical time, we had been in a position to determine much less regularly used C&C domains, which we suspect have been reserved for greater tier associates or extra essential campaigns.

Dynamic properties of Lumma Stealer

Lumma Stealer retrieves a dynamic configuration from the C&C server, which accommodates definitions specifying what to scan for exfiltration (see Desk 1). The first focus is on stealing net browser extension knowledge and databases containing passwords, session cookies, net searching historical past, and autofill knowledge. Apart from net browsers, it additionally focuses on stealing knowledge from password managers, VPNs, FTP purchasers, cloud providers, distant desktop functions, e mail purchasers, cryptocurrency wallets, and note-taking functions.

Desk 1. Dynamic config’s JSON fields

Although we haven’t seen important modifications within the default configurations, this characteristic enhances the malware’s capability to carry out focused exfiltration (see Determine 13). A complete overview of the configuration fields has already been properly documented on this analysis by SpyCloud.

The configuration is in JSON format, and it’s downloaded from the C&C server utilizing an HTTPS POST request that features the LID identifier, optionally available J parameter, and a particular hardcoded Consumer-Agent string.

The safety of the dynamic configuration has modified a couple of instances not too long ago. Previously, it was protected in the identical method because the static C&C listing, by a 32-byte XOR operate and base64 encoding. In March 2025 the safety modified to ChaCha20, the place the important thing and nonce had been prepended to the encrypted configuration.

The Consumer-Agent string is essential to observe, as offering it accurately is important for receiving the dynamic configuration. In April 2025, Lumma Stealer launched a further layer of obfuscation by encrypting JSON values utilizing an 8-byte XOR operate (see Determine 14).

This encrypted variant of the dynamic configuration is delivered when a barely up to date Consumer-Agent string is specified (see Desk 2).

Desk 2. Consumer-Agent variants

Apart from this dynamic configuration method, Lumma Stealer samples nonetheless include hardcoded directions for exfiltrating information. These embody knowledge from functions comparable to Outlook or Thunderbird, Steam account info, and Discord account tokens (see this SpyCloud blogpost). This mixture of dynamic and hardcoded configurations ensures that Lumma Stealer can successfully acquire a variety of invaluable knowledge.

To summarize all of the static and dynamic modifications talked about thus far, we’ve created a timeline (Determine 15) highlighting probably the most important developments noticed within the Lumma Stealer malware over the previous yr.

C&C communication

All through our Lumma Stealer monitoring interval, all extracted C&C domains persistently led to Cloudflare providers, that are utilized to hide Lumma Stealer’s actual C&C infrastructure. Cloudflare providers are additionally employed for C&C servers situated through dead-drop resolvers.

First, Lumma Stealer must receive an lively C&C server. The mechanism of this alternative is illustrated within the circulation chart proven in Determine 16.

Handshake

Though the precise handshake request to the C&C server will not be current within the newest Lumma Stealer builds, it’s price mentioning as a result of it was a characteristic of our monitoring for a very long time. The handshake request was an HTTPS POST request containing act=stay and a hardcoded Consumer-Agent. Energetic servers responded with a cleartext okay message.

Configuration request

When Lumma Stealer identifies an lively C&C server, it requests the configuration through an HTTPS POST request (Determine 17), which incorporates the LID and J parameters as knowledge. If the J parameter will not be current within the pattern, Lumma Stealer retrieves the default configuration from the C&C server. This configuration specifies what to scan for exfiltration, permitting the malware to adapt to completely different targets and environments.

Further payload execution

After Lumma Stealer efficiently exfiltrates delicate knowledge and harvested credentials, it points one remaining HTTPS POST request to the C&C server – this time, with a further sufferer {hardware} ID known as hwid. This remaining request retrieves a configuration of a further payload to be executed on the sufferer’s machine. The payload or a URL to obtain from is a part of that configuration. Notice that such a payload will not be at all times supplied.

Anti-analysis obfuscation strategies

Lumma Stealer employs a couple of, however efficient, anti-emulation strategies to make evaluation as sophisticated as doable. These strategies are designed to evade detection and hinder the efforts of safety analysts.

Oblique leap obfuscation

One of many major obfuscation strategies utilized by Lumma Stealer is oblique management circulation flattening, proven in Determine 18. This technique successfully disrupts the code blocks of the capabilities, making it practically inconceivable to maintain monitor of the operate logic. By flattening the management circulation, the malware obfuscates its operations, complicating the evaluation course of. For an in depth exploration of this system and thorough evaluation of those obfuscation patterns, together with a top level view of the answer, you possibly can confer with this complete article by Mandiant.

Stack strings

One other approach employed by Lumma Stealer is the usage of encrypted stack strings, as illustrated in Determine 19. This technique successfully hides binary knowledge and lots of essential strings within the Lumma Stealer pattern, making static evaluation of the binary tough. Furthermore, every encrypted string has its personal distinctive mathematical operate for decryption, including one other layer of complexity to the evaluation course of.

Import API obfuscation

In Lumma Stealer, imports are resolved at runtime. Import names are hashed utilizing the FNV-1a algorithm with every construct utilizing a customized offset foundation. As proven in Determine 20, since August 25^th, 2024, Lumma Stealer additionally obfuscates the FNV hash algorithm parameters through the use of stack strings.

Conclusion

This international disruption operation was made doable by our long-term monitoring of Lumma Stealer, which we’ve supplied an outline of on this blogpost. Now we have described the modus operandi of the Lumma Stealer group and its service. Moreover, we’ve documented the essential static identifiers and C&C communication in addition to its evolution during the last yr. Lastly, we summarized the important thing obfuscation strategies that make the evaluation of Lumma Stealer difficult.

The disruption operation, led by Microsoft, goals to grab all identified Lumma Stealer C&C domains, rendering Lumma Stealer’s exfiltration infrastructure nonfunctional. ESET will proceed to trace different infostealers whereas intently monitoring for Lumma Stealer exercise following this disruption operation.

For any inquiries about our analysis revealed on WeLiveSecurity, please contact us at threatintel@eset.com. 

ESET Analysis presents non-public APT intelligence experiences and knowledge feeds. For any inquiries about this service, go to the ESET Risk Intelligence web page.

IoCs

SHA-1	Filename	Detection	Description
6F94CFAABB19491F2B8E719D74AD032D4BEB3F29	AcroRd32.exe	Win32/Spy.Lumma Stealer.B	Lumma Stealer pattern – Construct 2024-06-27.
C5D3278284666863D7587F1B31B06F407C592AC4	Notion.exe	Win32/Spy.Lumma Stealer.B	Lumma Stealer pattern – Construct 2024-07-14.
5FA1EDC42ABB42D54D98FEE0D282DA453E200E99	explorer.exe	Win32/Spy.Lumma Stealer.B	Lumma Stealer pattern – Construct 2024-08-08.
0D744811CF41606DEB41596119EC7615FFEB0355	aspnet_regiis.exe	Win32/Spy.Lumma Stealer.B	Lumma Stealer pattern – Construct 2024-08-25.
2E3D4C2A7C68DE2DD31A8E0043D9CF7E7E20FDE1	nslookup.exe	Win32/Spy.Lumma Stealer.B	Lumma Stealer pattern – Construct 2024-09-20.
09734D99A278B3CF59FE82E96EE3019067AF2AC5	nslookup.exe	Win32/Spy.Lumma Stealer.B	Lumma Stealer pattern – Construct 2024-10-04.
1435D389C72A5855A5D6655D6299B4D7E78A0127	BitLockerToGo.exe	Win32/Spy.Lumma Stealer.B	Lumma Stealer pattern – Construct 2024-11-09.
2CCCEA9E1990D6BC7755CE5C3B9B0E4C9A8F0B59	exterior.exe	Win32/Spy.Lumma Stealer.B	Lumma Stealer pattern – Construct 2024-12-23.
658550E697D9499DB7821CBBBF59FFD39EB59053	Wemod-Premium-Unlocker-2025	MSIL/GenKryptik.HGWU	Lumma Stealer pattern – Construct 2025-01-18.
070A001AC12139CC1238017D795A2B43AC52770D	khykuQw.exe	Win32/Kryptik.HYUC	Lumma Stealer pattern – Construct 2025-02-27.
1FD806B1A0425340704F435CBF916B748801A387	Begin.exe	Win64/Injector.WR	Lumma Stealer pattern – Construct 2025-03-24.
F4840C887CAAFF0D5E073600AEC7C96099E32030	loader.exe	Win64/Kryptik.FAZ	Lumma Stealer pattern – Construct 2025-04-15.
8F58C4A16717176DFE3CD531C7E41BEF8CDF6CFE	Set-up.exe	Win32/Spy.Lumma Stealer.B	Lumma Stealer pattern – Construct 2025-04-23.

Community

MITRE ATT&CK strategies

This desk was constructed utilizing model 17 of the MITRE ATT&CK framework.