AI Security 101: Adopt AI in a secure and trustworthy way

Lotta Lundaas

Lotta Lundaas

VP of Marketing

September 26, 2023

min read

AI Security 101: Adopt AI in a secure and trustworthy way

Artificial Intelligence (AI) has become a cornerstone of modern business, driving innovation and efficiency across various sectors. However, as AI adoption increases, so do concerns about its security and trustworthiness. This blog post aims to shed light on the pressing security concerns in AI adoption and offers actionable insights to build trust in AI systems. 

The content is inspired by a panel discussion hosted by, featuring experts like Joseph Steinberg (Cybersecurity Expert & AI Advisor), Brett Norton (SVP Engineering,, and Yihua Liao (Head of AI Labs, Netskope).

The human element: a double-edged sword

As an AI-first company, we are confident that there is more good than bad when it comes to AI, but it’s important to be aware of the weaknesses in order to tackle them.

Joseph Steinberg, a cybersecurity expert, emphasizes that humans are both the strength and weakness when it comes to AI security. While AI systems offer tremendous benefits, they are programmed by humans who can make mistakes or even sabotage systems. Steinberg points out that if an AI system is fed bad data, it learns to make bad decisions. Therefore, the human element is crucial in AI security, and organizations must pay attention to who is programming their AI systems and what data is being fed into them.

The role of human error

Human errors can manifest in various ways, from coding mistakes to feeding incorrect data into AI systems. These errors can have far-reaching consequences, affecting the AI system's performance and, by extension, the organization's operations. Therefore, it's crucial to have a robust system in place to check and validate the data and algorithms that power AI systems.

Sabotage and malicious Intent

Another aspect of the human element is the potential for sabotage. Employees or external actors with malicious intent can manipulate AI systems, leading to adverse effects. Organizations must implement strict access controls and monitoring systems to detect and prevent such activities.

Deep fakes: A growing concern

Deep fakes are becoming a significant security concern in the AI landscape. These are AI-generated videos or audio recordings that are so convincing they can be used to impersonate individuals for malicious purposes. The technology behind deep fakes is advancing rapidly, making it increasingly challenging to distinguish between real and fake content. This poses a considerable risk and adds another layer of complexity to AI security.

The rise of deep fake technology

Deep fake technology has seen significant advancements in recent years. What started as a tool for creating realistic-looking images has now evolved into a technology capable of generating convincing videos and audio recordings. This rapid advancement poses a significant security risk, as deep fakes can be used for various malicious activities, including identity theft and disinformation campaigns.

Countermeasures against deep fakes

To counter the threat posed by deep fakes, organizations can employ AI-based detection systems that analyze videos and audio recordings to determine their authenticity. However, as deep fake technology continues to evolve, these detection systems must also be continually updated to stay effective.

Regulatory challenges

Joseph Steinberg highlights a significant hurdle in AI security: the lack of understanding among those responsible for creating regulations. According to Steinberg, the parties making the regulations often have little understanding of what they're regulating. This gap can lead to ineffective or even harmful regulations that fail to address the security concerns associated with AI adequately.

The need for industry standards

The regulatory landscape for AI is still in its infancy, and many countries are grappling with how to regulate this rapidly evolving technology. Organizations must stay updated on these developments to ensure their AI deployments are in line with current regulations.

One way to address the regulatory challenges is through the development of industry standards. These standards can serve as a guideline for organizations, helping them navigate the complex landscape of AI security.

Operational perspective: Partnering for security

AIIA reports that 81% of C-level executives rate AI a top priority or one of their top three priorities, but with many unknowns, how should business leaders tackle AI adoption? 

From an operational standpoint, Brett Norton suggests that small and medium-sized organizations should seek out partners experienced in AI and security. These partnerships can provide the necessary expertise and guidance to navigate the complex landscape of AI security. Norton likens AI security to other specialized fields, stating that just as you would consult a doctor for health concerns or a lawyer for legal issues, you should consult experts in AI security.

The importance of expert consultation

Expert consultation is crucial for navigating the complexities of AI security. As Steinberg and Norton both emphasize, AI security is a specialized field that requires a deep understanding of both AI and cybersecurity. Organizations should not attempt to navigate this complex landscape without expert guidance.

Building trust through rigorous testing and data encryption

According to a poll conducted during the panel discussion, the most important measures for building trust in AI systems are rigorous testing and validation of AI models (44%) and strengthening data encryption and access controls (33%). These measures are crucial for ensuring the secure and responsible deployment of AI technologies.

Rigorous testing and validation

Regular testing and validation of AI models are essential for building trust. These tests ensure that the AI system is performing as expected and that it is free from vulnerabilities that could be exploited by malicious actors.

Data encryption and access controls

Data encryption and robust access controls are other critical aspects of AI security. These measures protect the data that powers AI systems, ensuring that it cannot be accessed or manipulated by unauthorized individuals.

Final Takeaways

  • Human Involvement: Pay close attention to who is programming your AI and what data they are using.
  • Expert Consultation: Consult with experts in AI and security to navigate the complex landscape.
  • Regulatory Awareness: Stay updated on the evolving regulatory framework and align your AI deployments accordingly.
  • Rigorous Testing: Regularly test and validate your AI models to ensure they are secure and reliable.
  • Data Encryption: Strengthen data encryption and access controls to protect sensitive information.

Building trust in AI is not just a technological challenge but also a human and organizational one. By understanding the security risks and taking proactive steps to mitigate them, organizations can responsibly adopt AI technologies that offer tremendous benefits. As Joseph Steinberg aptly puts it, "An ounce of prevention is worth many tons of cure."

For more insights on AI adoption and security, feel free to reach out to us at, where we are pushing the boundaries of what's possible with autonomous finance and accounting through AI.

This blog post is inspired by a panel discussion hosted by, featuring experts in the field of AI and security. Click to watch the full recording.

Be a Finance Pioneer

Sign up to receive the latest news, updates and valuable accounting and finance resources conveniently in your inbox.