New Aokigahara Malware Targets ChatGPT, Claude, Codex, and Cursor Login Credentials

A newly observed Windows infostealer called Aokigahara is targeting login credentials and session data associated with ChatGPT Desktop, Claude, Codex, and Cursor.

The malware searches infected computers for authentication files and application storage, extending familiar password theft into the AI tools developers increasingly use for everyday work.

The threat is not evidence that these AI services have been breached. Instead, attackers collect locally stored information that may help them access accounts.

Similar AI coding agent theft has shown why tokens, saved connections, and project records deserve the same protection as browser passwords and cloud keys.

Researchers from Check Point identified Aokigahara in the wild and detailed its capabilities in an October 23 research report.

New Aokigahara Malware

They noted that an earlier version may have been known as Kamidere or Stygian, while stressing that the operators’ methods for making money from stolen information remain unconfirmed.

A dedicated collection module named aisessions searches for data linked to Claude Code, Claude Desktop, Codex, ChatGPT Desktop, and Cursor.

Its scope also covers Windsurf, GitHub Copilot, Gemini CLI, JetBrains AI Assistant, Amazon Q, and Kiro. The module additionally queries Windows Credential Manager for stored credential information.

For Claude, collection targets include credential files and desktop application folders. Codex targets include authentication, configuration, and environment files.

Cursor’s local storage databases are also selected, while ChatGPT Desktop’s application cache is searched. These locations may contain useful secrets, but not every targeted file necessarily holds a password or token.

Earlier reporting on Aoki credential stealing described similar interest in AI application data. The practical risk depends on whether collected secrets remain valid and reusable: working session tokens could give attackers account access without requiring them to enter the victim’s password again.

That access could expose private conversations, uploaded material, or development information. Attackers might also consume paid services or resell compromised accounts.

Check Point described these as potential outcomes, not confirmed activity by Aokigahara’s operators, and raised possible data extortion where stolen access reveals sensitive business content.

A separate ide module widens collection to Git credentials, SSH files that may contain private keys, and AWS, Azure, and Google Cloud configuration data.

Alongside its AI focus, Aokigahara steals browser credentials and cookies, captures screenshots and clipboard contents, and collects wallet, messaging, VPN, and email application data.

Evasion, Encrypted Uploads, and Incident Response

Aokigahara includes optional checks designed to frustrate malware analysis. Depending on configuration, it examines debugger indicators, virtual machine traces, hardware resources, process counts, and mouse movement.

It can also stop running when keyboard or Windows interface language settings match an excluded group, including Russian, Ukrainian, and several regional languages.

Its configuration is encrypted with AES-256-CBC and controls which built-in collection modules execute. It also stores server details, communication settings, and fallback destinations.

Importantly, disabled collection modules remain embedded in the executable, so the capabilities present in a sample may exceed those enabled during a particular run.

The malware packages collected information into ZIP archives, then encrypts uploads using AES-256-GCM. An X25519 key exchange helps establish the encryption key.

Separate requests open an upload session, transfer each collection category, and mark completion. The latest analyzed samples use a Windows-looking User-Agent string, although that string alone cannot prove infection.

Upload headers identify the collection category and sequence number, allowing the server to separate browser data from AI sessions and other stolen material.

The final completion message signals that the upload phase ended; it does not independently confirm that every stolen archive reached the server.

The wider trend is already visible in incidents involving a fake Gemini npm package that targeted local data from several AI tools.

That separate campaign does not establish Aokigahara’s delivery method; it shows how developer workflows have become attractive places to collect reusable access credentials.

For defenders, the collection pattern supports monitoring unexpected processes that read AI authentication stores, editor databases, cloud credentials, and SSH material.

Network findings should be checked alongside endpoint evidence rather than treated as standalone proof. Check Point also published a YARA rule to support identification of the malware.

After a suspected infection, responders should isolate the device and use a clean system to revoke AI sessions, rotate exposed keys, and review connected cloud and repository accounts.

Resetting passwords alone may leave stolen sessions usable. Teams should also review AI account activity and determine whether sensitive conversations or uploaded files were accessible.

The findings concern theft from compromised endpoints, not a demonstrated vulnerability in ChatGPT, Claude, Codex, or Cursor.

Organizations should therefore include AI application sessions in their existing infostealer response plans rather than treating them as ordinary application settings.

Indicators of compromise (IoCs):-

Type	Indicator	Description
IP address and port	2[.]26[.]252[.]48:8080	Aokigahara command-and-control infrastructure
Domain	dfwioeiofwr[.]info	Aokigahara command-and-control infrastructure
Domain	dobeqnesmot[.]cc	Aokigahara command-and-control infrastructure
Subdomain	gate[.]dfwioeiofwr[.]info	Aokigahara command-and-control infrastructure
Domain	hjazfioe[.]surf	Aokigahara command-and-control infrastructure
Domain	kkeyyou[.]cc	Aokigahara command-and-control infrastructure
Domain	neihite[.]cc	Aokigahara command-and-control infrastructure
Domain	somelitersofblood[.]pw	Aokigahara command-and-control infrastructure
SHA-1	8d6c86e0af7807f9b5baf15e9f076bb81e012515	Dropped clipper component

The following table reproduces all 150 Aokigahara sample SHA-1 hashes. Each cell represents a separate sample; columns do not indicate relationships between samples.

Aokigahara SHA-1	Aokigahara SHA-1	Aokigahara SHA-1
007137599bc4d69ee952a5228b5fcb2c3b9a95a7	00d52c7e759d878f0d8a117502f6d854af13f007	04eddccb79db41e03a4ee8542e8a33d7f1d9fe18
057a2d1dedeb93b7765c455a9e0c469e7cbe4843	079ddf5cc7d6ca2e02f860f87d4d18c83db9d331	0b74ce90a39278fb97fe94f6306af71cbe679503
0c46ac5555cf8f305c1345a21d0802c52bf52cd8	0d8d0bad2e08856ba379e425e2e90e49506af631	0e76150cee3f2b89f0fbe39106ab2898db109f09
0f69792cb926a416766d47a30a05009c13be0d64	130f2710cf9863bdd6724e8912600505b6fb7ef2	133dc00ad81638e6b0a2f8f163b4d694ce477165
151d8013ca3ab3dab9cd37f4ed9870cc83e87a77	1701db4fa798b309cb25227408406dde4c69de8a	1835f1e0b1eda3dc4b39682ecad3eba88cbbc44e
18e041b5a0c668392ff4d036f3f9660c2698311e	1a6a6e08a01cc3f1455af2c4ecd59a3d7802cd7a	1a7248720f3ac79082841b006582f20799b314a9
1a97a18db74f2b6df984c9a188a79b4e65a06b81	1cb6ccfdcb277a0277729443d15f17a441fa0eda	1cb8dc11bdb8baf342b83079fb650cd6bb1b132d
1e072236f87cd914dc583ca2163460d656bfcfba	1e94d7f1d501724076fcf98459e14b68321d45f7	2045f165d737f2d43ba603d444dfa5d9e6f05805
211e45f306b84a82cfe736635c176a424e7479dc	21ef070e39f651118a804882374dc2dfa3e38666	2362a326070d83c167251873b389bd783781cf6c
241e37b8fc5ffeefbd1d89a82b74cf2d92b49eac	244371fc1642f0548b807b0a4a2e7ea5d7b4c82a	28099fc5a8cbbd40173f8ac0aded9044fdb4a598
2cdc77108c585d14b1567053b8f8a58ea78c255f	2f30b116ca3dc48ff5783c3bb7391d671866e27b	30737fa46f2ff93e33f45008cf306233954ea898
324e0fffc7977f8ff591993819123d33c2970953	3586eeaf3ffe2c3a4f1d005aee863b7aee7cc7bd	3696baa239c479f9f5103618214972890f1a473b
385e21a96687a33f79f9cab0627aec7218e1a8f4	38b2c29b46086b417181deb958e8b276a9d1fbf4	3aaeb49864108c92af50feeb5d9e3b3efaa39818
3c56a0c3c59a1425bfdcdffda9e8b308f478b147	426bea548a24f7f39c279770b5dd21fbed7f7328	4417d2645adb6e7e94959daf483f93058500840f
4485feb1385975ef13f08148cdc679f50fa0cc6d	456414c0acf463e06f95f5bd3d8355758fca6b53	4595d1875441ffb8c076856e366d1f4f5c74a679
48c6b66dc4b4d40d72b7460e6fd670b6c03d5d40	4b3a4ad0acd43fc8796147381074a98e8e576904	4c07021f8b0cdd3ca334a13f9af2bade9e699dc2
50060c0fc5fa8932671eb24b03b05d00c2e99951	51ae6cad54f33846f83a352457b9a44ef8b7a61d	59bbaca8a848215e7e4af7cec320432c51b0b812
5ef7eb755f1e3f6bf8586794777466a639057dcb	5fadf206dc163f63ebf66f984042b00c67121813	5fe11fbb89d33efa712ffb8590626bd88fecab8f
61a024f5a42a9a91b0a9871a1613c7fb3ef5613b	62dcc9ba2a5eee9dd941ad08766603c89a45869b	6393c0d3fb76f771a9418c2d152c621929ee002b
63cc5033532cd00bda3d4968b1257629e613505a	682dd6392004e1131b1aa3f2351a0bc7adb84d14	699c809a116106da40b3f641c11b0db34d0f1c5e
6a09fffca381f6624d2def1083284ca486306e7c	6c860007f8f49de8e9848b538bd7acdc6bfcdf55	6e073a196a9bfa01da418b5783c2395fb22c7ec9
6eb8f17dba69f6d25b7f5c529b25d199707b69e2	6f7d5935602c188b1e3066b4ab9aeef7b85a8ec2	70915fac07af4ad7acc1868fe05dc00b98206f0f
70d4007b11e56752fffb81f71c90987f4bb441fa	70e91644802053cfa68be6ec0f94207282ff27e4	7233a15327dc2f0cca0292e6352eebf6e6e090d8
72aa6417a19c45dabdd6c79087fb13c73ae7b823	7cc6c75fad93b2482a9c5f4522530007c72f99e0	82ebfc09c98bd211ff8b08efcd8ec8b9ce1fb4be
83b43ce0ee9881201451c0604fb5b792bd4581f5	86230c07e638b85bc8010df1aa9e33ca61fda4cc	89f296aef7988bdd59109e4263d0a31c886770aa
8b0cf62382f620bbbf5d176295d63ed522ab9c2c	8c3111a61fbf30e86b8206397ce3c0017e5ea509	8c4c1d8325adb463c34694ce3a5a851e97b81a6f
8d7a81119ce7be6eb452192f08fa2669228ff597	8d860ca64ce0cbf40871fd66b9e3a513e73cda90	8dc40b7208b8c0d95d1ab8dda9a56fff790b3ddd
8e9fd1274b77ff9726c4637d008a9d599c9d5f1e	9003310bf73dc337b8b50131f1c0724a79161aac	90c057dafefbee423e3014266218272d454a0aa3
912ed8ce7767f94f62de6a30fb8e9df7c0c512f7	9369e0e5eae22cff71b375ad285ea89471f25101	97c80e2d382e6e55f61702da3afb838fdec2f4f1
98c6649cc256dfadf1af939da46f037b05c0f2e5	99b28a20851a3ed29769a469f4dd31ebbb8a83de	9d424ce23753ffddc7b39477222123b0f2c33066
9f05045f5447d79c8c0d0f5cb589fd85c89620d3	9f20034441e78574aec7ad39842a342f939eaf6a	a00da33342a5ea5168d851a5c93109e07cf95c49
a2abdf1af38da9995a3b6672ce6c34e9e17c39a4	a4871f4a61faf74e1cceaffe26901f6076c9f37e	a63ded811d412cd0f859891b0c1f066b87bb1b80
a84148b4d408544325c20ee04299ae95c762b6a2	ad960d2abb1999a7260c5f498532a5bb8754c2a5	ae9d6b1a32a45b05edb920538e538a3d6a5fe439
af196456d2ebf9556ab239661922565e1bca9ac0	b335483800d35a29c25767c917725e24c21b0f2b	b33e59c598845d2547ddc7f654115df2284bb3ae
b36d0796e33f808eaeb33b33908327a410680c56	b4491f1919a88deedebe7fba56b593e7d6655458	b523d56d82485253f2e3619437c8398b7ee0f20d
b607025d3e06d1e8cffcd398c64871d7fcd92bf4	b6f0dfd3aa2fc7b8b2ad432c4a4276da23ceb46b	bab4c910e3c98085bf228519f346d071a75e44e3
bad339c5ad65808285b0e87498f2ea6f669417d9	c00aceaa40b38e48f662ac927bb86302934aaf96	c2e28ba5523c1f3a545a0b4a00a9d5e9c8ae7583
c323c59815082ac842f0bafcc01f537444e1cc5a	c5b50b5e14dfb5e028dfd88a4e4fb960dd650b52	c810acdf30bcdfc6e8021e7f62a3bcce0828c2ee
c8456bb71a74ed6305455c318295b8a39e9277f3	cbdff0df0dbdbc3c1845d208661dd75fd335ef93	cc12fcb9e5e6f7afde66f5c3452ed6e3cd6d5390
cc708ae3d7aa084279122c0db7250096f750d6b0	d34491da5aa159314129c0ff67f957d873703e35	d3c90275dd7279a791d717676be15789376f7aa7
d8bfff5139b54b6f98f07c051cd25d7aa465fec8	da986a3cb2ce67456ef374fbdd957d370a1a4367	daef439e1f110d624ce50fe0f9d9d24f61f7497b
db3332ce75676389b79cf3603d81cb231cb7c19c	db69c7c930fd5dc5097847979a965fdea4171d5b	dba5fef27b53ff479a4d7046cd1f22710081a16c
de9d152010819c8fb5cb611a4de920399e69cfeb	dfd1540c0185e74ec65acfcae1af7ba0da36c774	e052ff49190f54e0ffc52f7b3f7b5e7b9e121c4c
e3af6d9e3b35e5e04feda3cde96256dd3b099624	e4d93fa1b2d0fd26e8b877511ea33badb4628169	e5c43abd1c212b2f2a61d5e1cb7821406ca5dc8e
e8d09eb2e5857fde39b44aeb48078f5396809026	ec7cd90d56e5b73caa9d4eb0472bfee68d716121	eec7ece53d47e50940584f4dec684a24a78f42f2
f0b23f420659babb3a730b122259440091c53c12	f30e9730db0663f87c67a4b9a0cf12ee59b1621a	f35e6d2936c0f6dc26bdc68f6d1599cf438a863a
f4184e1bb61e19d190555c50fdb9d4591767bf53	f47d45eea1c42b066f8537baf72ffa266646bf62	f5993e64a18e7494b25a185de9f63a1df961b99a
f633fc9fd16dadca2725d3d54aaf13ea1a62174f	f6d950a5a8e5ba336f748b66745c6c1f189d3362	f8cf33b18e9932d285696fee40a7fee2a87341c2
f8dbc552c89bef758e4fcbd9942d06bc0c11db0a	f990bf4f05dfa4d86b3e3bc0a54f62e91b83f2c6	fa61cf3bebefd88ea9d06007a4b85791aa851c8a
fbe8708247fabd8a1bc3963db3911b5e77728669	fe6b7d6fa8c2424d14a0d465ab78aaf83db0d07b	ff402e64bc42222c8a4ee703123903c2a2117dd9

Additional source-reported network and file artifacts can support behavioral hunting. Unlike the listed sample hashes and command-and-control destinations, these artifacts are not necessarily malicious in isolation.

Artifact type	Source-reported value	Context
HTTP endpoint	/gate/handshake	Establishes an upload session
HTTP endpoint	/gate/chunk	Receives encrypted stolen-data archives
HTTP endpoint	/gate/complete	Receives the completion notification
HTTP User-Agent	Microsoft-CryptoAPI/10.0	Used by the latest analyzed samples
HTTP headers	X-Build-ID, X-Gate-Token, X-Session, X-HWID, X-KEX, X-Chunk, X-Chunk-Index	Communication and upload metadata
Upload category	aisessions	Identifies collected AI application data
Configuration section	.stgcfg	Stores encrypted malware configuration
Configuration marker	STGC	Identifies the encrypted configuration structure
Component directory	C:MicrosoftVault	Location beneath which the clipper component is dropped
YARA strings	AKGRES, AISessions/, aokigahara, kamidere	Strings included in Check Point’s detection rule

Note: IP addresses and domains are intentionally defanged (e.g., [.]) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms such as MISP, VirusTotal, or your SIEM.

Give your SOC the alert context to stops incidents quickly. Integrate Threat Intel from 16K SOCs.

The post New Aokigahara Malware Targets ChatGPT, Claude, Codex, and Cursor Login Credentials appeared first on Cyber Security News.