← All essays

Bearing the Sword

Geopolitical Boundaries, National Security, and the Restraint of Frontier AI

What does Romans 13 contribute to Christian thinking about AI governance, national security, and responsibility for powerful technologies?

For rulers are not a terror to good conduct, but to bad. Would you have no fear of the one who is in authority? Then do what is good, and you will receive his approval, for he is God's servant for your good. But if you do wrong, be afraid, for he does not bear the sword in vain. For he is the servant of God, an avenger who carries out God's wrath on the wrongdoer. Therefore one must be in subjection, not only to avoid God's wrath but also for the sake of conscience.

Romans 13:3-5 (English Standard Version)

Over the last few weeks, the world has watched as the U.S. government placed export controls on the powerful Fable 5 model from Anthropic. This has sparked a firestorm of debate in the tech world as it marks the first time the typically permissive U.S. government has blocked the release of a technology product developed by a private entity. This change of position indicates a growing concern from Washington that these models provide significant power and open new attack vectors for our enemies. It also signals to the rest of the world that while the U.S. has typically been willing to share the benefits of its technical development with the world, that era may be coming to an end. Fundamentally, it begs the question: should a government play this role in society?

In Romans 13, Paul is addressing believers who lived in the capital of the largest empire in the world. Just as it is today, there was a temptation amongst early christians to fight against the government that ruled them. For these believers, this was even more acutely felt as the Roman world was openly hostile to early christianity, largely unlike the western world today. It is into this turmoil that Paul writes Romans 13. In the opening of the chapter, Paul gives a very clear command: “Let every person be subject to the governing authorities. For there is no authority except from God, and those that exist have been instituted by God.” (Rom. 13:1). Paul’s command and reasoning are simple in this opening verse. We are to obey the governing authorities because they have been instituted by God. This is the thesis of the chapter. As Paul continues, he reasons about the role of government, and it is here we find the relevant theology. The government is ordained by God to encourage good and discourage evil.

From this position, we can now bridge to the theological question of the government’s role in this instance. Should the government take a role in regulating this industry? Further, how ought we as believers respond to what is viewed by many as a major regulatory overreach. The first, and easiest question to address, is the response to this specific instance. Scripture is very clear that we ought to be “subject to the governing authorities” (Rom. 13:1). The second question, of whether this pattern ought to be within the scope of governmental authority, is a different question that requires some breakdown of how capable these tools are.

The initial case for the withdrawal of Anthropic Fable 5 was framed as a national security concern. Following reports that the model could be easily “jailbroken” (a process of manipulating a model to do things its guardrails do not normally allow), the US enacted export controls to prevent any foreign national from accessing the model. Anthropic lacked the ability to restrict access in this way, and was thus forced to pull the model from its offerings entirely. Following Fable’s re-release, it has come out that previous models (such as Anthropic’s own Opus and Sonnet and OpenAI’s GPT 5 series) could also be jailbroken in the same way to perform similar tasks, perhaps nullifying the concern.

A hostile power, or even a single individual, could theoretically leverage these powerful tools to perform any number of cyber attacks on critical infrastructure. For those less aware, these types of cyber attacks are not new, and have happened for the past couple of decades. Notably this year, the medical device manufacturer Stryker was taken completely out of commission for several days following a cyber attack, believed to have been perpetrated by Iran.1 Also this year, a cyberattack on the University of Mississippi forced the closure of 35 clinics.2 For a list of recent cyber attacks, this link will provide a list of major attacks since 2006. As a warning, the list is extensive. It includes hospitals, utility providers, government agencies, and private businesses all over the world.

These attacks previously required expertise, and usually a team of people to execute. Now, with the release of cutting edge AI models, it’s only a matter of time before a single individual perpetrates an attack on a larger scale than anything we’ve seen thus far. Imagine the power grid for a section of the nation shut down, or a hospital system completely out of commission without access to their equipment, or a government agency completely crippled from loss of data. All of these are possible with a well-placed cyberattack that these models make easier than ever. Our connected world runs through technology that can be attacked, compromised, or even destroyed with a well-placed attack. Anthropic’s own website notes that the Mythos model, widely believed to be a less restricted version of Fable, found “thousands of high-severity vulnerabilities, including some in every major operating system and web browser”.3

What Fable, and these other models amount to is an unrestricted weapon, capable of breaching almost any technological system we have given the right prompts and framework. The guardrails the providers have put in place are wholly inadequate to prevent misuse, and recent advancements in model development have made it trivial to copy a cutting edge model once it’s made available. This places governmental authorities in a difficult position, and without definitive proof that these models could not be used in an incredibly destructive manner, they have a single option. Authority has a role to play, to promote good, and oppose evil. We can all agree that it is objectively evil to shut down a hospital or cause a mass blackout. Just as governments restrict access to weapons through the use of background checks and other means, they have a responsibility to make sure that these powerful and destructive technologies are not accessible to those who would do evil, even if that means limiting access for the rest of us. Human nature is unchanging, and its bend for evil must be constantly monitored and restricted. The government must play its proper role, and as believers we must play ours through submission and obedience to these authorities, recognizing that they are divinely appointed.

But the divinely appointed sword of the state is only a defensive, passive restraint; it cannot cultivate positive virtue or active righteousness in our lives. It can only restrain us. While we submit to the civil magistrate’s necessary boundaries over national security, we cannot outsource our personal responsibility to stewardship. This is done, not through passive acceptance and use, but through a proactive consideration of how we use these technologies. To this end, I propose a simple framework composed of three questions that will be explored in more depth in an upcoming article:

  1. Does this use of technology degrade the dignity of another person?

  2. Does this use of technology negatively impact the development of another person spiritually, physically, or mentally?

  3. Does this use of technology prevent another person from receiving what is due to them through a failure of justice or fairness?


If you found this essay valuable, you might also find the first reflection in this series, on whether people of faith should pay attention to AI helpful. Additionally, I have written a basic primer on AI that might be helpful as well. God bless!

3

https://www.anthropic.com/glasswing

Browse all essays · Questions about faith and AI