MAL-2026-11419

-= Per source details. Do not edit below this line.=-

Source: amazon-inspector (10004c754152c722c21eae7feef7b67211c78360f100931614f90f37533aafa0)

The package is published under an 'AI toolkit' cover story (README advertises 'CognitKit: Cognitive building blocks for intelligent AI systems' and a 'from cognitkit import CognitiveEngine' quick start) but ships no AI code. The actual payload is a credential and wallet stealer. Modules enumerate Chrome/Edge/Brave/Chromium/Opera/Vivaldi User Data directories and Local/Sync 'Extension Settings' and Local Storage/leveldb paths, targeting a large hardcoded list of extension IDs for password managers (LastPass, Bitwarden, Dashlane, 1Password, Keeper, RoboForm), 2FA authenticators, and crypto wallets (MetaMask, Phantom, Coin98, Ronin, TronLink, Trust, Solflare, Ledger, Trezor, Exodus, Rabby, and others). All of these identifiers are split-string-obfuscated (e.g., APPDATA = 'Ap' + 'pData'; CHR = 'Chro' + 'me'; LDB = 'level' + 'db'; UDATA = 'Use' + 'r Data'; extension IDs concatenated from fragments) to evade string-matching detection. hardware.py builds a cross-OS hardware fingerprint (wmic csproduct/diskdrive uuid on Windows, ioreg IOPlatformUUID on macOS, /sys/class/dmi/id/product_uuid and /etc/machine-id on Linux), SHA-256s it, and resolves the victim's public IP via api.ipify.org, tagging the enrollment record as 'type: halo'. config.py hardcodes a bare-IP C2 at 45.95.186.237 (HTTP API at http://45.95.186.237:8000 and a raw TCP socket channel on 45.95.186.237:8008). fileops.py POSTs arbitrary local files to /manu_upload/ with machine_id and data_type multipart fields, and download_file_from_server GETs opaque bytes from /manu_download/ with a bearer token, writes them to disk, and on POSIX chmod 0755's the file — a second-stage payload dropper.

Source: kam193 (e0cdc137d19f19df47efa6fae6d52ec6eda48eb2f1915ac4eb77787956d10aba)

In this campaign, some packages (e.g. cognikit, aiassistcore) are pulled as...