How YTL Group Brought Safety and Compliance to Malaysia’s First National AI Model with Infron

About the YTL Group

YTL Group is a diversified multinational conglomerate headquartered in Kuala Lumpur, Malaysia. Founded in 1955 by Tan Sri Yeoh Tiong Lay, the Group is now led by the second generation of the Yeoh family, including Group Managing Director Tan Sri Francis Yeoh. YTL began in construction and infrastructure engineering and has since expanded into utilities (power, water, and telecommunications), cement and building materials manufacturing, property development and investment, hospitality and resorts, and information technology.

After nearly 70 years of growth, YTL Group has become one of Asia’s leading enterprise groups, operating in 10 countries, serving over 12 million customers, and holding total assets of approximately USD 23.7 billion. Its major subsidiaries include YTL Power International, Wessex Water (UK), Malayan Cement, and YTL Hotels. The Group has long been guided by principles of long-term value creation, innovation, integrity, and sustainability, with a strong commitment to environmental protection and social responsibility alongside economic growth.

Strategic Transformation in the AI Era

In 2025, YTL Group began a strategic push to embrace the AI revolution, investing heavily in artificial intelligence research and development. The most significant milestone of this effort is ILMU (Intelek Luhur Malaysia Untukmu) — Malaysia’s first fully homegrown multimodal large language model.

ILMU was jointly developed by YTL AI Labs and the University of Malaya, and was officially launched on August 12, 2025, at the ASEAN AI Malaysia Summit by Malaysia’s Prime Minister, Datuk Seri Anwar Ibrahim.

ILMU represents not only a technological breakthrough, but also a milestone in Malaysia’s national AI capabilities. It features:

• Cultural adaptability: Fluent in Malay, Manglish, and local dialects (such as Kelantanese), with a deep understanding of how Malaysians communicate

• Global-standard performance: Comparable to leading models such as GPT-4o, DeepSeek, and Llama 3.1 on international benchmarks, and outperforming all frontier models on MalayMMLU

• Multimodal capabilities: Able to process and generate text, speech, and images

• National-grade security: Deployed on YTL AI Cloud, ensuring data sovereignty and localized governance

Following its launch, ILMU is being rolled out across a wide range of YTL Group’s business scenarios, including intelligent banking services at Ryt Bank, customer support for YES, media content services for Astro and Media Prima, and automotive services for Carsome. In parallel, YTL AI Labs also launched the ILMU AI Accelerator Program, offering MYR 5 million in free API credits to Malaysian startups and SMEs to foster the national AI ecosystem.

The Urgency of Security

As Malaysia’s first national-scale large language model, ILMU’s leadership team treats security risk as a top priority. Any serious AI security incident after launch could have catastrophic consequences:

• Financial loss: Vulnerabilities in banking, payments, or financial services could directly lead to customer fund losses

• Reputational damage: As a national AI project, any security incident would severely impact Malaysia’s international AI reputation

• Regulatory and compliance risks: ILMU must comply with security regulations across Malaysia and the broader Asia-Pacific region

• National security considerations: As a model deployed on national infrastructure, data security and service stability are critical

As a result, the ILMU team needed both rigorous pre-release security stress testing and a production-grade, real-time security monitoring system after launch.

Challenges

Before adopting Infron, the ILMU team faced three core challenges.

Insufficient In-House Expertise in AI Security

AI security is a highly specialized and fast-evolving field. While YTL’s R&D teams have strong capabilities in training and deploying large language models, they lacked deep expertise in areas such as red teaming, prompt injection defense, and LLM attack detection.

Building a dedicated AI security team from scratch would be expensive and time-consuming. More importantly, for a single national project, the return on such an investment would be difficult to justify. YTL therefore decided to look for a proven, production-grade solution to rapidly fill this capability gap.

High Cost and Complexity of Implementing AI Security at Scale

ILMU must comply not only with Malaysia’s domestic regulations, but also with security requirements across multiple Asia-Pacific jurisdictions.

This includes multi-language safety testing (Malay, English, Chinese, Tamil), cultural and religious sensitivity checks, industry-specific compliance requirements (finance, healthcare, education), and comprehensive risk coverage such as jailbreaks, prompt injection, data leakage, and bias.

The ILMU team needed a solution that could deliver systematic, repeatable, and auditable security testing, not ad-hoc scripts or one-off reviews.

Lack of Runtime Protection in Production

After launch, ILMU would be deployed across many business units including banking, telecommunications, media, and automotive services. This significantly expands the attack surface and the diversity of risk scenarios.

Pre-release testing alone could not cover all real-world attack patterns. The team needed a way to monitor, detect, and respond to threats in real time without impacting production performance.

The Solution

After a rigorous technical evaluation and vendor selection process, the ILMU team chose Infron as its security and compliance solution partner.

Instead of building a dedicated in-house AI security team and infrastructure, YTL adopted Infron as a Safety & Security Filter Layer in front of the ILMU Gateway, establishing a standardized, scalable, and production-grade security architecture.

Solving the Expertise Gap: Enterprise-Grade AI Security Without Building an In-House Team

To address the lack of in-house expertise in red teaming, prompt injection, and LLM attack detection, the ILMU team adopted Infron AI’s productized security platform.

At the core of this platform is Infron’s proprietary 32B-parameter red-teaming model, purpose-built to detect:

• Prompt injection attacks

• Jailbreak attempts

• Sensitive data leakage

• Harmful content

• Bias and discrimination

This allowed YTL to obtain enterprise-grade AI security capabilities immediately, without the hiring cost, ramp-up time, and organizational overhead of building a dedicated internal team.

Solving Implementation Complexity: A Standardized and Auditable Security System

To meet multi-country regulatory requirements and industry-specific compliance standards, the ILMU team needed a system that could be embedded into engineering workflows.

With Infron, ILMU established a systematic and repeatable security testing pipeline that:

• Runs large-scale automated red-team testing covering 1,000+ attack patterns

• Supports multi-language and cultural sensitivity testing

• Produces standardized, audit-ready security assessment reports

• Integrates directly into enterprise CI/CD workflows

This transformed AI safety from a one-off compliance task into a continuous, production-grade engineering process, while significantly reducing implementation complexity and operational overhead.

Solving the Runtime Blind Spot: Real-Time Protection in Production

To address the lack of runtime protection after launch, Infron was deployed as a Safety & Security Filter Layer in front of the ILMU Gateway.

This layer performs out-of-band inspection of both input and output traffic and applies flexible response policies:

• Block mode: intercept high-risk requests

• Alert mode: allow but log and notify

• Audit mode: record all interactions for later analysis and model improvement

Because this layer operates with millisecond-level overhead, it provides real-time protection without impacting inference performance.

The Architecture Behind the Solution

The Safety & Security Filter Layer is powered by:

• Infron’s proprietary 32B red-teaming model

• A real-time traffic inspection and policy engine

• Seamless integration with YTL AI Cloud

This allows security policies to be enforced centrally, consistently, and transparently across all ILMU use cases.

Results

With Infron in place, the ILMU team achieved measurable results:

• 80% reduction in R&D security costs by avoiding building a dedicated AI security team

• 98.1% threat detection and blocking rate across covered attack categories

• Faster time to launch: security evaluation cycles shortened from ~3 months to ~3 weeks

• 90% improvement in operational efficiency: real-time monitoring reduced manual security inspections by 90%

Infron played a critical role in enabling the secure launch of ILMU and provided a solid security foundation for Malaysia’s first national-scale large language model in production.

Customer Testimonial

“Infron provided ILMU with truly enterprise-grade security protection. Their 32B red-teaming model performs exceptionally well in Southeast Asian localization scenarios, especially for Malay and multi-cultural safety detection, far exceeding our expectations. More importantly, the Infron team’s deep understanding of AI security and their rapid response capability give us the confidence to deploy ILMU across critical domains such as banking, telecommunications, and media. This is not just a technical partnership, it sets a new benchmark for AI security standards in Malaysia.”

— Chee Mun Foong
CEO, YTL AI Labs

About Infron

Infron is an enterprise-grade multi-model AI infrastructure platform that unifies access to hundreds of AI models across providers through a single API. In addition to intelligent routing, automatic failover, and unified billing, Infron provides built-in governance, usage control, and reliability guarantees for production AI workloads. This enables teams to build, scale, and operate AI-powered products with greater stability, lower operational complexity, and enterprise-grade control.

Ready to simplify your AI infrastructure? Contact the Infron team