TamperedChef (EvilAI) Activity Clusters — Trojanized Productivity Apps Deliver RATs/Stealers via Malvertising; 4,000+ Samples, 81 Code-Signing Orgs Tracked (Unit 42)
TamperedChef (EvilAI) Activity Clusters (TL-2026-0541), also tracked as TamperedChef, is a high-severity malware campaign, first published 2026-05-20. It is attributed to TamperedChef with medium confidence, affects Microsoft Windows, maps to 29 MITRE ATT&CK techniques (T1005, T1027, T1036), and is covered by 9 detection rules and 30 indicators of compromise.
Key facts for TL-2026-0541
- Threat ID
- TL-2026-0541
- Also known as
- TamperedChef, EvilAI, ManualFinder, OneStart Browser, AppSuite PDF, Calendaromatic
- Severity
- HIGH
- Status
- ACTIVE
- Category
- MALWARE
- First published
- 2026-05-20
- Last reviewed
- 2026-05-20
- Attribution
- TamperedChef
- Attribution confidence
- MEDIUM
- Motivation
- FINANCIAL
- Target sectors
- technology, education, financial, healthcare, retail, media, general_consumer
- Target regions
- North America, Europe, Asia-Pacific, Latin America
- Detection rules
- 9
- Indicators of compromise
- 30
Malware and tooling in TamperedChef (EvilAI) Activity Clusters
Malware and tooling: AppSuite PDF Editor, AsyncRAT, Calendaromatic, CrystalPDF, EvilAI, JustAskJacky, Lumma Stealer - S1213, ManualFinder, OneStart, Stealc, TamperedChef
Palo Alto Unit 42 documented three overlapping activity clusters (CL-CRI-1089, CL-UNK-1090, CL-UNK-1110) corresponding to the public TamperedChef / EvilAI threat. The operation distributes trojanized productivity software — PDF editors, calendars, ZIP extractors, AI helpers — through malvertising and professionally built lookalike sites; binaries are Authenticode-signed by short-lived shell corporations, rebuilt every 1–4 weeks to defeat hash detection, and remain dormant for weeks-to-months before fetching secondary payloads such as infostealers, RATs, and residential-proxy clients. Unit 42 tracked 4,000+ samples, ~100 unique variants, and 81 code-signing organizations across the clusters, with named campaigns including AppSuite PDF, Calendaromatic, JustAskJacky, and CrystalPDF.
How TamperedChef (EvilAI) Activity Clusters works
## Overview
TamperedChef (also tracked as EvilAI, ManualFinder, OneStart) is a long-running, financially-motivated malware-distribution operation that has been active since at least mid-2023. Palo Alto Unit 42's May 2026 report consolidated the public ecosystem into three Unit 42 activity clusters — CL-CRI-1089, CL-UNK-1090, CL-UNK-1110 — based on overlaps in code-signing certificates, Electron application boilerplate, install-time scheduled tasks, and a shared post-install command-and-control choreography. The cluster naming is operational; multiple vendors (Truesec, G DATA, Field Effect, Expel) have published overlapping subsets of the same activity under different brand names.
## Distribution and Social Engineering
The operators run thousands of professionally designed lookalike websites for fictitious productivity utilities — "AppSuite PDF Editor," "Calendaromatic," "JustAskJacky," "CrystalPDF," "PDF OneStart," "OneStart Browser," "ManualFinder" — each with full marketing pages, fake review carousels, support docs, and an EULA. Traffic is driven by Google Search ads (malvertising) and SEO poisoning against terms like "merge PDF," "calendar app," "free PDF editor," and "AI assistant." Installer bundles are MSI or NSIS packages built around an Electron shell that ships a real (or convincingly real) productivity feature, so victims experience a working application and do not immediately suspect compromise.
## Code Signing and Rebuild Cadence
A defining tradecraft trait is heavy abuse of legitimate Authenticode signing. Unit 42 enumerated 81 distinct code-signing organizations — most are LLCs registered in low-friction jurisdictions (UK Companies House, Panama, US states with anonymous registration) that exist only long enough to obtain an OV code-signing certificate, sign a wave of binaries, and burn before revocation. Certificates rotate every 1–4 weeks; binaries are recompiled with cosmetic changes between rebuilds to invalidate every hash-based reputation lookup. Roughly 100 distinct variants observed across the 4,000+ samples reflect template families rather than truly different malware.
## Execution Chain
1. **Initial Access.** Victim clicks a sponsored search result or social-media ad, lands on a polished product page, downloads an MSI/NSIS installer. 2. **Install-time Persistence.** The installer drops the Electron application into `%LOCALAPPDATA%\Programs\<AppName>\` and registers a scheduled task (e.g., `\Microsoft\Windows\<AppName>Update`, `\<AppName>SchedulerTask`) that runs the main binary with a benign-looking flag at logon and every 1–6 hours. 3. **Dormancy.** The first executions perform no malicious activity. Telemetry is minimal — pings to a benign-looking update endpoint to confirm reachability. Dormancy windows of 14–90 days are typical; some samples wait until day-60 to flip. 4. **Activation.** A backend logic flag (date-gated and/or geo-gated) triggers the JavaScript runtime in the Electron app to load a remote payload chunk over HTTPS. Common dispatch URLs use Cloudflare-fronted custom domains and rotating Cloudflare Workers endpoints. The payload is staged as an `.asar`-style archive or a Node `require()`-able blob and executed inside the Electron process to avoid spawning a fresh PE. 5. **Secondary Payloads.** Observed final-stage families include AsyncRAT, Lumma Stealer, StealC, RedLine variants, AnubisStealer, residential-proxy clients (Honeygain-style SDKs repackaged), and bespoke crypto-wallet sweepers. Some samples additionally enroll the host into a click-fraud / ad-fraud botnet. 6. **Discovery and Collection.** Stealers harvest browser cookies, saved passwords (Chromium login data), crypto wallet files (`wallet.dat`, MetaMask localStorage), Discord and Telegram tokens, FTP/SSH client configs, and screenshots. 7. **Exfiltration.** Data is exfiltrated over HTTPS to actor-controlled C2 servers, typically with simple POST of a ZIP archive or chunked JSON.
## C2 Infrastructure
C2 domains follow a recognizable pattern: short, English-word "product" names registered through Namecheap or Porkbun, fronted by Cloudflare, and pointing to VPS instances at low-reputation hosters (Stark Industries, FBW Networks, AS200019, IPXO ranges). TLS certificates are Let's Encrypt or Cloudflare Edge; SAN entries frequently include multiple unrelated "product" hostnames hinting at shared infrastructure across campaign families. Beaconing is HTTPS POST/GET with JSON bodies, jittered randomly 30–300 seconds during active mode and once-per-6-hours during dormancy. No custom protocol — the choice of Electron and JavaScript makes the network behavior identical to any benign Electron app's auto-updater.
## Attribution
Unit 42 attributes the activity to a financially motivated criminal ecosystem rather than a single actor; the certificate-issuance and infrastructure-rotation discipline is consistent with an organized affiliate-style program. No nation-state attribution. Some overlap exists with the operators behind the OneStart browser PUP family and the ManualFinder install-base from 2024.
## Defender Implications
Because every sample is freshly signed and freshly hashed, hash, certificate-thumbprint, and reputation-only controls fail. Behavioral detection on (a) MSI/NSIS installers dropping Electron apps into `%LOCALAPPDATA%\Programs`, (b) scheduled tasks created at install time that execute Electron apps with non-default args, (c) Electron renderer processes making outbound HTTPS POSTs to non-vendor domains after dormancy, and (d) Chromium login-data file reads by non-browser processes are the most durable signals. Application allowlisting and code-signing-issuer allowlisting (rather than block-known-bad) materially reduce exposure.
MITRE ATT&CK techniques used in TL-2026-0541
Collection
T1005 Data from Local System; T1113 Screen Capture
Defense Evasion
T1027 Obfuscated Files or Information; T1036 Masquerading; T1497 Virtualization/Sandbox Evasion; T1564 Hide Artifacts
Exfiltration
T1041 Exfiltration Over C2 Channel
Persistence
T1053 Scheduled Task/Job; T1547 Boot or Logon Autostart Execution
Execution
T1059 Command and Scripting Interpreter; T1569 System Services
Command and Control
T1071 Application Layer Protocol; T1090 Proxy; T1102 Web Service; T1105 Ingress Tool Transfer; T1573 Encrypted Channel
Discovery
T1082 System Information Discovery; T1083 File and Directory Discovery; T1518 Software Discovery
Initial Access
execution
Impact
Credential Access
T1539 Steal Web Session Cookie; T1555 Credentials from Password Stores
defense-impairment
Resource Development
T1583 Acquire Infrastructure; T1587 Develop Capabilities; T1588 Obtain Capabilities; T1608 Stage Capabilities
Affected products and versions in TamperedChef (EvilAI) Activity Clusters
- Microsoft — Windows
Vulnerable versions: 10; 11; Server 2019; Server 2022 - Generic — Chromium-based browsers (credential exposure target)
Vulnerable versions: Chrome; Edge; Brave; Opera
Remediation for TamperedChef (EvilAI) Activity Clusters
Immediate actions
- Block known C2 domains and IP ranges at the perimeter (see IOC list)
- Hunt for scheduled tasks under \Microsoft\Windows\ that execute binaries from %LOCALAPPDATA%\Programs\ with Electron-style arguments
- Enumerate Authenticode-signed binaries in user profile paths whose signer is on the Unit 42 81-org list and quarantine pending review
- Search EDR for child processes of Electron-shell binaries spawning powershell.exe, cmd.exe, or accessing Chromium Login Data
Workarounds
- Disable user-writable scheduled task creation via Group Policy where feasible
- Browser policy: disable installation of unknown extensions and force SafeBrowsing Enhanced mode
Longer-term hardening
- Deploy application allowlisting (Windows Defender Application Control or AppLocker) with publisher allowlist limited to known-good ISVs
- Restrict standard users from installing MSI/NSIS packages via Group Policy (InstallElevated, DisableMSI)
- Block sponsored search-ad downloads at the secure web gateway and force download inspection for executable types
- Implement EDR rule: alert on outbound HTTPS POST from Electron child renderer processes to domains < 90 days old
Weaknesses (CWE) in TamperedChef (EvilAI) Activity Clusters
CWE-506, CWE-829, CWE-494
Timeline of TamperedChef (EvilAI) Activity Clusters
- Earliest samples of the OneStart browser PUP family — the genealogical ancestor of the TamperedChef ecosystem — appear in vendor telemetry.
- Expel documents OneStart pivoting from PUP-style adware behavior to delivery of stealer payloads, marking the operational shift to malware distribution.
- Truesec publishes the first detailed write-up of "TamperedChef" — a malicious PDF editor campaign delivering an infostealer via Google Ads.
- BleepingComputer amplifies the TamperedChef finding, raising operator-side awareness and accelerating certificate-rotation cadence.
- Field Effect publishes infection-chain analysis linking TamperedChef to the broader ManualFinder install base, expanding the known footprint.
- G DATA publishes "EvilAI — when AI helpers turn into malware," documenting fake AI productivity apps as an additional cluster lure.
- Microsoft Threat Intelligence highlights code-signing-certificate abuse by malvertising operators, naming overlapping families.
- Palo Alto Unit 42 publishes the consolidated cluster report (CL-CRI-1089, CL-UNK-1090, CL-UNK-1110): 4,000+ samples, ~100 variants, 81 code-signing organizations tracked.
- As of 2026-05-29, TamperedChef/EvilAI remains an active, financially-motivated malvertising operation: Unit 42's 2026-05-20 cluster report and Sophos/Truesec/G DATA confirm ongoing activity with new components surfacing. Some domains went dark and one cert wave was revoked, but operators keep rotating infrastructure and signing certs, so it is contained-but-live, not neutralized.
Sources cited for TamperedChef (EvilAI) Activity Clusters
- Tracking TamperedChef Clusters via Certificate and Code Reuse
- TamperedChef: Malicious PDF Editor Campaign
- G DATA: EvilAI — when AI helpers turn into malware
- Field Effect: ManualFinder and TamperedChef infection chain
- Expel: OneStart PUP-to-malware pivot
- BleepingComputer: Fake PDF editor pushes information-stealing malware
- Microsoft Threat Intelligence: Code-signing abuse by malvertising operators
- MITRE ATT&CK: Code Signing (T1553.002)
Threats related to TamperedChef (EvilAI) Activity Clusters
- SourTrade Malvertising: ServiceWorker-Orchestrated In-Browser Assembly Builds a Unique Windows Executable Per Victim
- P2P Botnets in the Wild: Pink, Hajime, Mozi, FritzFrog, and Panchan — Decentralized C2 Landscape (360 Netlab Continuous Monitoring)
- Indirect Setup-Error Prompt Abuse: Clean GitHub Repo + Failing Python Package + DNS TXT Payload Tricks AI Coding Agents (Claude Code) into Running a Reverse Shell
Detection coverage for TL-2026-0541
As of 2026-05-20, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-0541 across Splunk SPL, Microsoft KQL and Sigma, covering 30 indicator(s) of compromise. The whole corpus is readable without an account; a free account unlocks full detection query text in Splunk SPL, Microsoft KQL and Sigma; paid tiers add raw indicator values, correlation and the MCP server. Threadlinqs MCP server · View plans.