Is your GenAI adoption outpacing your ability to secure it?

Generative artificial intelligence (GenAI) continues to soar in Asia Pacific, with spending predicted to reach US$26 billion by 2027. This highlights organizations’ strong belief in GenAI’s potential to enhance product development and design workflows, automate processes, and generate content, ultimately creating better business outcomes.

Central to this growth is the evolving capabilities of large language models (LLMs). These advances, such as the ability to process multiple data types like text and images, as well as provide richer context, can offer more accurate insights, improved support for different applications, and increased efficiency.

Innovating at the cost of security risks

But as with any technological advancements, GenAI comes with risks.

Early adopters leveraging LLMs, such as the open-source model DeepSeek for its high performance in reasoning tasks, may be capturing business value ahead of their peers. However, Gartner cautioned that the swift adoption of GenAI technologies has outpaced the development of data governance and security measures.

“At present, there is a growing number of LLM vulnerabilities that can potentially expose businesses to threat actors. Among the more prominent ones is the Grandma Attack, a social engineering attack in the form of prompt injection that jailbreaks the safeguards of LLMs with specific inputs. In this instance, threat actors craft prompts that mimic the persona of a harmless elderly relative to execute specific instructions,” explains Ker Yang Tong, ASEAN and India CTO of Fujitsu. “This allows them to bypass security controls to extract sensitive data and even manipulate critical infrastructure controls, resulting in severe financial and reputational damage.”

Another risk involves bypassing content filters. Threat actors can generate factually incorrect or inappropriate information to spread disinformation or conduct phishing campaigns. These LLM-related risks are often made more complex by existing data silos within enterprises, and an increasingly fragmented cybersecurity landscape fraught with sophisticated cyberattacks.

To highlight how the meteoric rise of LLMs raises critical safety concerns, Fujitsu used its LLM vulnerability scanner to conduct the most extensive security analysis of DeepSeek’s flagship model, DeepSeek-R1 7B, to date, surpassing others in scope and attack coverage. Through comprehensive testing and over 7,000 simulated attacks, the scanner revealed that DeepSeek-R1 is the worst performing LLM model against malware and phishing attacks, with a 100% attack success rate in bypassing the model’s safeguards.

A proactive security approach to GenAI

While 45% of CIOs surveyed by IDC emphasized security as their primary concern for GenAI initiatives, it’s concerning that only 22.4% of organizations felt adequately prepared for AI-ready trust and security. With so much at stake, enterprises must prioritize a security-first AI strategy, such that security and governance are implemented at the same pace as innovation.

“In this new era of AI, CIOs must adopt a long-term vision for innovation. True AI isn’t a race to the finish line; it’s a strategic tool that should be leveraged via a security-first lens,” says Tong. “This will not only allow businesses to drive business continuity, but also avoid pitfalls that may erode public trust.”

Tools such as the Fujitsu LLM vulnerability scanner help enterprises to adopt a proactive cybersecurity stance. Popular LLM models—from DeepSeek-R1 to Llama 3.1—were analyzed, and more than 7,700 attacks were conducted, spanning 25 distinct attack types.

By leveraging a database that aggregates state-of-the-art information, including LLM attack scenarios and vulnerabilities published by academia and the AI security community, as well as Fujitsu’s proprietary techniques and the latest attack techniques, the scanner provided unprecedented visibility into an LLM’s attack surface. This empowers enterprises to take a risk-based approach to AI adoption, prioritizing security without stifling innovation.

Understanding and mitigating the security risks of LLMs is central to reaping the full benefits of GenAI. As GenAI continues to evolve, a security-first approach that can provide comprehensive visibility into the threat landscape will be paramount to business success.

Find out how Fujitsu can help maximize the business value of your GenAI strategy today.