APT36 Uses AI-Generated “Vibeware” Malware Against India

Pakistan-linked threat actor APT36, better known as Transparent Tribe, has changed how it builds malware. In a report published in March 2026, researchers at Bitdefender describe a campaign against Indian government networks in which the group used AI coding tools to mass-produce large numbers of cheap, disposable implants — an approach Bitdefender calls “vibeware”, a play on “vibe coding”.

Who Is APT36?

APT36 (also tracked as Transparent Tribe and Mythic Leopard) has been active for more than a decade and is associated with Pakistani state interests. Its traditional targets are the Indian government, military and diplomatic missions, typically reached through spear-phishing with defence- or government-themed lures. Bitdefender links the new activity to the group with medium confidence.

Rows of blinking server drives glow under red emergency lights as holographic data streams float through a dusty, high-contrast room, illustrating the urgent vulnerability of digital infrastructure to advanced AI threats.

What “Vibeware” Means

Vibeware is not a new kind of social engineering and has nothing to do with voice cloning. It describes AI-assisted malware industrialization: instead of investing in a few carefully engineered tools, the attackers used AI assistants to generate many small implants quickly, in a wide range of programming languages — including Nim, Zig, Crystal, Rust, Go and C#. Less common languages such as Nim, Zig and Crystal make signature-based detection harder, because security tools have far fewer reference samples for them.

Bitdefender describes the goal as a “Distributed Denial of Detection”: flooding defenders’ telemetry with many different, short-lived binaries rather than achieving technical sophistication. Named implants include Warcode, NimShellcodeLoader, CreepDropper, SHEETCREEP, MAILCREEP, SupaServ, LuminousStealer, CrystalShell, ZigShell and others.

Targets

Rows of black server racks with blue lights stand before a curved wall displaying a digital map of India, illustrating the nation's secure data center infrastructure and critical cyber defense capabilities.

The campaign targeted the Indian government and Indian embassies abroad, as well as the Afghan government and some private companies. According to the researchers, the attackers used LinkedIn to identify high-value individuals before sending them lures.

How the Attack Unfolds

A cybersecurity analyst works late in a dim office, surrounded by monitors showing code and network graphs, reflecting intense focus on defending against evolving digital threats.

Infection typically starts with a phishing email carrying a ZIP or ISO archive that contains a Windows shortcut (.LNK) file, or a PDF — for example a fake résumé — with a “Download Document” button that leads to an attacker-controlled site. PowerShell scripts then load the implant in memory. For command and control, the malware hides in legitimate cloud services: Google Sheets, Google Drive, Slack, Discord, Supabase, Firebase and the Microsoft Graph API. Because this traffic goes to trusted platforms, it is difficult to block at the network perimeter.

Quantity over Quality

The AI-generated tools were far from perfect. Bitdefender found many of them “unstable and riddled with logical errors”; in one case an implant lacked a working server address to send stolen data to. The danger lies in volume, not polish: even a modest success rate pays off when producing a new variant costs almost nothing.

A digital shield protecting a glowing globe against streams of code fragments, symbolizing defense against high-volume malware campaigns.

Defense and Mitigation

Because individual samples change constantly, defenders should focus on behaviour rather than signatures: block or closely monitor .LNK, ISO and script attachments, restrict PowerShell with constrained language mode and logging, and use EDR tools that flag unusual child processes. Outbound connections from workstations to services such as Google Sheets, Supabase or Firebase that are not used by the business deserve scrutiny. User awareness remains important too — especially for staff whose roles are visible on LinkedIn.

Conclusion

APT36’s vibeware campaign shows a realistic, near-term effect of generative AI on cyber espionage: not super-intelligent malware, but cheap mass production that tries to overwhelm detection. For government security teams, the answer is behaviour-based detection and tight control over the scripts and cloud services their endpoints are allowed to use.

Sources: The Hacker News, Mar 2026; Hackread: APT36 floods Indian government networks with AI-made vibeware