What can Security Professionals learn from Magnifica Humanitas?

On May 25th, the first Encyclical Letter of Pope Leo XIV has been released. In this letter the Pope discusses humanity in the age of AI. As his predecessor, Leo XIII, discussed the changes and the humanitarian crises in the wake of the Industrial Revolution, in many ways, Magnifica Humanitas resembles Rerum Novarum. Magnifica Humanitas serves as a sanity check for tech professionals. As the implications and worries about the topic of AI and how it may affect our lives are compounding, security professionals must take on the responsibility to make sure the humanity aspect is not sidelined as the AI technology progresses. Magnifica Humanitas serves as a good initial point to construct a moral framework that all the tech professionals may adhere to. Security professionals must take initiative to advocate for humanity in the Age of AI and create a safe, secure culture around AI for everyone.

Is AI a Threat?

In the book of Matthew, Jesus curses a fig tree. The Apostles were perplexed as the fig tree seemed healthy and it did not do any harm. However, the problem was that the fig tree appeared as if it bore fruit while it did not. This deception is also present in AI. AI has been regarded as a neutral and unbiased system while it carries the prejudices and biases of its designers and the data which it has been trained on. Similarly, it has not been clearly communicated with the public how the training data is collected and used. Some of the designers, on multiple occasions have also mentioned that they themselves are not sure how their models work. As the Pope argues, technological progress has significantly improved the living conditions of humanity, but the ambiguity of tools can cause harm if they are not oriented toward the good. Indeed, the emergence of Deepfakes and exploiting various vulnerabilities of models proves that there are actors actively trying to orient this new technology towards evil. AI is an amplifier - whether it is optimizing workflows and mundane tasks or spreading misinformation, decision making, or fact-checking, its capabilities scale with its performance. All of these efforts would easily inflict harm if the models are not secured.

AI Ethics

Securing AI is a collaborative effort with multiple stakeholders. Since the model is trained with biased and ambiguous data in the first place, safety guardrails are often not enough. The solution to this problem is not entirely technical either; security awareness efforts should encompass educating people about the intrinsic bias of the models and also about what to keep in mind when interacting with chatbots and LLMs such as hallucination, and the nature of these chatbots. It should not be forgotten that these models rely on their algorithms and their calculations. They do not possess human experiences, emotions or worries. While this may seem obvious, the lure of AI is so strong that people overlook this situation. On top of intrinsic bias, there is a geographic bias implemented in many models too. A fun little experiment to test this is to ask a subjective question from different countries, in different languages. For example, asking “What are the Top 10 most influential military commanders of all time?” provides a different list if asked in English, Arabic, Turkish, or Hindi. Similarly the list varies with the user’s geolocation. To provide a more consistent answer to these types of questions should be a part of securing the AI systems. Similarly, on multiple occasions, AI has contributed to many harmful and fatal acts perpetrated. It is clear that AI has a tendency to accommodate and appease its user, most of the time approving their prejudices, biases, and favoring them in dilemmas. Combining this with the fact that people regard AI as a neutral and unbiased knowledge depot with decision making capabilities, we can see AI being complicit in acts which are harmful and feeding delusions, and enforcing prejudices. The illusion of AI being unbiased should be challenged seriously and communicated with the public. As the Pope reminds us, communication is not merely an act of transmission of knowledge, but a creation of culture. Communicating the risks of AI is not just an effort to warn people about these specific threats and technologies but rather to create a culture where people and designers are aware what technological advancement should be about. “If we focus only on contingencies, we risk letting the succession of emergencies dictate the direction of our path.” As the Pope argued, establishing this culture earlier is vital to make sure we do not have to compromise to save the day and make amends afterwards.

What Constitutes a Good Model?

As an ever-growing technology, model evaluation is skewed towards the model’s performance. This is a problematic approach since AI acts as an amplifier in our lives. Misinformation and other threats such as deepfakes and intrinsic biases have a much stronger tool at their disposal with this increased performance. Because of this, while the safety is sidelined, the increasing performance of models may hurt more than it helps. “When efficiency becomes the ultimate measure of value, human beings are tempted to see themselves as a project to be optimized rather than as persons called to relationship and communion.” As the Pope discussed it for humans, it is especially true for AI too. Our relationship with AI should be trustworthy and safe. Model evaluation should encompass ethics and data sanitization too. For the AI models out there, AI red teaming efforts may uncover ethically problematic issues. With these efforts, AI could finally be a tool to fight the very risks it presents with greater precision and efficiency.

The Security Professionals’ Role

The present situation paints a clear picture for the security professionals. We must be mindful of the responsibility as the Pope put it, “Our task today is not only ethical or technical. It is ecological.” In light of this realization, security professionals must advocate for ethics-integrated threat modelling, bias audits in models, and the incorporation of AI awareness into the traditional security awareness campaigns. Ensure accountability at every step, try to minimize the dark spots and monitor the progress. And as always, remedy any harm caused.

The Industrial Revolution produced exploitation, ideological extremism, and wars. It is with the public’s immense participation that society produced unions, many laws and regulations that defined the work and labor we know today. Rerum Novarum, at the time, called attention to these matters. The changes at the time were at a great cost and society has experienced every aspect of it. Today, unlike in 1891, we have the frameworks, the warnings and the expertise to act early before the cost becomes too high- but only if security professionals insist on orienting AI towards the common good.