5 tips: how to navigate security challenges in AI tools

  • 02/10/2023

Artificial Intelligence (AI) is rapidly growing in the manufacturing industry, and for good reason. It has the potential to increase productivity, and efficiency, and even move businesses ahead of competitors. However, as we all know, with great power comes great responsibility. You must ensure your organization is equipped to safely develop and use AI tools without putting your organization at risk.

Developing your own AI solutions gives you full control over its security whereas for existing AI tools, the responsibility for security often lies with the vendor. Nonetheless, you don’t want to put your organization at risk so it’s important to have a good view of the security aspect of the AI tool you are already using or would like to use in the future.

Therefore, we’ll share with you essential tips on how to do this, while avoiding any potential pitfalls.

Julie Asselbergh

Information Security Officer

Data protection - safeguard your valuable assets

Data is the lifeblood of any AI tool. In this case, your company data. Ensuring its protection is absolutely crucial for many reasons such as strengthening customer and stakeholder trust. Furthermore, your data protection policy must comply with relevant privacy laws and regulations such as GDPR, CCPA, and soon the AI Act.

It’s important to have clear processes on how data is collected, stored, used, and disposed of. Every company needs to implement top-notch cybersecurity measures to keep its data safe from breaches and other potential threats. Encrypt sensitive data and ensure secure communication channels are used between your AI systems and other software components. In addition, it is essential to have mechanisms in place for data subjects to exercise their rights, such as requesting a copy of their data or requesting deletion.

While open-source AI tools can be extremely useful and cost-effective, they also come with potential risks. It is crucial to thoroughly research and evaluate the tool's history, including reading reviews from other users, and ensuring that it has a strong track record of security and privacy practices before incorporating it into your organization's AI framework. When it comes to tools like ChatGPT, it is important to ask questions about their data privacy policy. Does ChatGPT keep private information about its users? This question becomes even more relevant if you are entering data about your clients into the system, as it may require you to inform your clients about how their data will be used. Always exercise caution and avoid uploading or entering sensitive or personal information into tools whose data practices are not transparent and well-documented.

This information is easy to find if you’re looking into Microsoft’s AI tools. The most accessible tool is Copilot. Powered by Azure OpenAI Service, Copilot offers a combination of responsible AI principles and Azure's industry-leading security. Integrated into services such as Dynamics 365 and Microsoft 365, it inherits the robust security, compliance and privacy practices of these systems. With features such as two-factor authentication and privacy protection, Copilot positions itself as a trusted AI platform.

What sets Copilot apart from other tools is the unique way it generates AI responses. These are specifically contextual, directly relevant to the task at hand, and informed by your business data. You can have a look at one of our previous articles for some examples. It’s designed with a strong focus on security and privacy:

  • Your data will never be used to train Large Language Models (LLMs).
  • You can withdraw your approval for Copilot’s access to Microsoft 365 Graph and Dataverse at any given time.
  • The response will never include information to which the user does not have access unless explicitly configured otherwise.

In the case of critical applications, it might be a balanced strategy for many organizations to combine existing AI tools with custom development. Using Microsoft’s Azure AI Platform provides you with several built-in security features.

Restricted access - keep your AI systems safe

Restricting access to your AI systems is imperative to ensure that only authorized personnel can make changes. Implement role-based access controls (RBAC) so that employees can only interact with the AI systems within the boundaries of their job description. This means that a data analyst, for example, would not have access to system settings and a system administrator might not have access to raw data. This not only keeps your systems and data safe from potential internal threats and human error but also helps maintain accountability, as specific actions can be traced back to individual employees.

Microsoft facilitates setting up Access Control within your tenant. Their permissions models help ensure that the right groups and users have access only to the data they’re supposed to have access to. Here as well, all prompts and responses are saved, so there’s a full audit trail for investigators.

Regular updates - stay one step ahead

Just as with any other software, it's important to keep your AI systems up-to-date. Software and systems are constantly being improved and at the same time, new vulnerabilities are being discovered. Regular updates will help safeguard your AI against these vulnerabilities and stay ahead of potential issues. Make sure to establish a schedule for AI systems and related software updates and closely monitor them to ensure that your systems are protected. It's not just about updating, but also regularly checking for patches and security enhancements released by software vendors. This way, you minimize the chances of malicious actors exploiting these vulnerabilities.


Training and awareness - empower your workforce

Technological security is only as strong as the weakest link, and often that link is human. That’s why you should regularly provide AI safety training sessions to your teams. As a well-informed workforce, they will understand the potential risks and how to mitigate them. They’ll be better equipped to tackle problems when they arise. Encourage an open dialogue within your organization about AI safety, and keep your employees updated on the latest threats and prevention techniques.

There are plenty of relevant sources to find content for such training. Microsoft’s AI learning and community hub is only one example.

Create a guideline on AI usage in your organization

Many of the topics touched upon above can have their place in a corporate guideline for AI usage. Such a guideline should be a framework for the responsible use of AI within your organization.

  • Include all stakeholders in the process of setting this up
  • Roles & responsibilities
  • Approved tools and evaluating new tools
  • IP rights
  • It should be a living document

If you’re looking for inspiration, you could have a look at Microsoft. They defined six ethical principles that they feel are essential to creating responsible and trustworthy AI as it moves into mainstream products and services. These are:

  • Fairness: How might an AI system allocate opportunities, resources, or information in ways that are fair to the humans who use it?
  • Reliability and Safety: How might the system function well for people across different use conditions and contexts, including ones it was not originally intended for?
  • Privacy and Security: How might the system be designed to support privacy and security?
  • Inclusiveness: How might the system be designed to be inclusive of people of all abilities?
  • Transparency: How might people misunderstand, misuse, or incorrectly estimate the capabilities of the system?

Conclusion

To summarize, it is crucial for businesses to recognize the significant data implications of AI and establish comprehensive guidelines before embarking on AI implementation, whether utilizing internally developed tools or third-party solutions. Additionally, assigning appropriate user rights, facilitating skill development, and conducting regular updates are essential steps in ensuring secure and effective AI usage. By prioritizing these measures, companies can proactively mitigate security risks and maximize the benefits of AI technologies.

AI Act

A flagship initiative to regulate AI based on its capacity to harm. Ranging from low to unacceptable risk, the requirements for the AI application will be different.

Dataverse

Securely stores and manages data that are used by business applications.

Microsoft 365 Graph

Provides a unified way to access and manipulate your business data, enabling you to build apps that can interact with Microsoft’s cloud services.

Stay up to date!