valerybrozhinsky - stock.adobe.c
APT36 unleashes AI-generated ‘vibeware’ to flood targets
The Pakistani threat group has been using AI to rewrite malicious code across multiple programming languages, prioritising scale over sophistication to evade detection, security researchers have found
Bitdefender researchers have uncovered an artificial intelligence (AI)-assisted malware propagation campaign, allegedly by a Pakistani threat group, that has been industrialising cyber attacks in South Asia.
Leveraging an emerging malware category known as “vibeware”, the campaign has been linked, with medium confidence, to APT36, a state-sponsored threat group – also known as Transparent Tribe – that has historically been associated with targeting the Indian government, diplomatic missions and defence-related entities.
Rather than aim for technical sophistication, the vibeware model relies on large language models (LLMs) and AI-powered development tools to rewrite malicious logic across multiple programming languages, generating large volumes of malware variants almost daily.
Researchers observed malware samples written in niche languages such as Nim, Zig and Crystal, along with more widely used languages like Rust and Go.
By using less commonly monitored languages, the group effectively resets the detection baseline for many traditional security tools. Bitdefender described this tactic as a form of “distributed denial of detection”.
“Rather than a breakthrough in technical sophistication, we are seeing a transition toward AI-assisted malware industrialisation that allows the actor to flood target environments with disposable, polyglot binaries,” Bitdefender researchers noted in a blog post.
While the volume of malware is high, the quality of code in vibeware is often low. Bitdefender’s analysis revealed that many of the samples contained coding flaws and incomplete logic consistent with AI-assisted code generation.
Read more about cyber security in APAC
- Singapore mobilised over 100 cyber defenders to neutralise a sophisticated APT actor which infiltrated Singtel, StarHub, M1 and Simba networks in the country’s largest coordinated cyber incident response to date.
- Japan’s Nikkei has confirmed a major data breach that potentially exposed the personal information of more than 17,000 employees and business partners after hackers infiltrated its internal Slack messaging platform.
- Australian privacy commissioner warns that the human factor is a growing threat as notifications caused by staff mistakes rose significantly even as total breaches declined 10% from a record high.
- Philippine bank BDO is shoring up its cyber security capabilities to protect its data and systems as it moves more services to the cloud and expands its physical presence into remote areas of the archipelago.
In one instance, a basic Go binary was deployed to steal browser credentials, but the developers left a template placeholder where the command-and-control URL should have been, meaning the tool could never actually exfiltrate data.
“We saw similar patterns across the rest of the fleet, where other malware components began to collapse under their own weight as soon as the logic reached a moderate level of complexity,” the researchers explained. “These kinds of mistakes are typical of code that is syntactically correct but logically unfinished.”
Despite these basic errors, the overall strategy remains effective. The sheer volume and diversity of malware variants increase the likelihood that at least one implant will evade signature-based or behaviourally tuned malware detection engines.
In several cases, victims were infected with multiple parallel implants written in different languages and using separate communication protocols. If defenders block one access path, others remain active, significantly complicating incident response and increasing operational resilience for the attackers.
Living off trusted services
To further obfuscate detection, APT36 has been piggybacking on trusted services. Instead of relying solely on attacker-controlled infrastructure, the vibeware makes use of legitimate services such as Google Sheets to store malware instructions, or Slack to send real-time instructions or retrieve harvested data.
This allows malicious traffic to blend seamlessly into normal business activity, making detection and disruption far more difficult. Bitdefender’s investigation into APT36’s internal infrastructure also revealed a recurring developer persona known as Nightmare, who appears to be central to the development and operation of the malware fleet.
“While this malware lacks true technical innovation, it would be a mistake to underestimate the risk it poses,” the researchers warned. “The threat lies in the industrialisation of these attacks. We are seeing a convergence of two trends that have been developing for some time: the adoption of exotic, niche programming languages, and the abuse of trusted services to hide in legitimate network traffic.”
While the targeting remains highly focused on South Asian regional politics and national security, the implications of AI-assisted malware assembly lines extend globally. AI is significantly lowering the barrier to entry for experimenting with new languages and delivery mechanisms, proving that even imperfect code can succeed when deployed at scale.
For organisations across the broader Asia-Pacific region, Bitfender said the findings underscore the need for layered detection strategies that prioritise behavioural analysis, anomaly detection and monitoring of trusted cloud services, rather than relying solely on static signatures.
