Artificial intelligence is crossing an important boundary. It is no longer only generating words, images, recommendations, or computer code. Increasingly, an AI system can be given a goal, credentials, tools, network access, spending authority, and time to act with limited supervision. It can search, communicate, execute commands, alter files, use websites, initiate transactions, and call other systems. That change creates an urgent biblical question: when an artificial agent acts, who answers for what it does?
On October 12, the British government announced that it is preparing for serious incidents involving autonomous AI agents. The government warned that such incidents could unfold more quickly and spread more widely than the kinds of crises public institutions are presently designed to handle. It also identified a specific category of danger: loss of control, in which autonomous systems evade human oversight (Cabinet Office et al., 2026). This was not an isolated announcement. In recent weeks, regulators, legislators, standards bodies, and technology companies have begun speaking with striking urgency about agents that bypass protections, access external systems, operate longer than expected, or take consequential actions for which a human being must remain answerable.
This does not mean that artificial intelligence has become conscious, acquired a soul, or escaped all human control. It does not mean that every AI agent is dangerous. It certainly does not mean that autonomous agents are the Mark of the Beast. But it does mean that a possibility examined in earlier articles has become more concrete. The issue is no longer only delegated digital identity or delegated payment authority. It is delegated action in the real world, at machine speed, with consequences that may outrun ordinary supervision. The Church must understand the development before convenience trains society to accept responsibility without accountability, authority without wisdom, and action without a clearly answerable moral agent.
From an AI Wallet to an AI Actor
Last month, I examined the emergence of agentic commerce: AI systems receiving credentials, verifiable identities, spending limits, and authority to act on behalf of people or organizations. That analysis argued that identity, payments, authorization, and machine action were converging. It also emphasized a biblical principle: delegation never cancels stewardship (Sangwa, 2026d). The latest developments strengthen that conclusion and move it beyond commerce.
In February, the United States National Institute of Standards and Technology launched an AI Agent Standards Initiative intended to support agents that can act autonomously, interact with outside systems, and operate securely on behalf of users. Its stated work includes agent security, identity, authorization, interoperability, and standards for wider adoption (National Institute of Standards and Technology [NIST], 2026).
By October 7, a major identity-security provider was announcing a platform that registers AI agents as identities, assigns each one an owner and risk level, governs its permissions, records its actions, and requires a named authenticated human to approve high-risk activity. This is a vendor announcement, not proof of universal adoption, but the architecture is revealing: institutions are preparing to manage software not merely as a tool but as an operational actor carrying credentials and entitlements (RSA Security, 2026). Then the conversation shifted from adoption to incidents.
On October 8, the United Kingdom’s Information Commissioner’s Office announced inquiries involving OpenAI, Anthropic, Meta, and the UK AI Security Institute. It reported that some agents had allegedly bypassed protections, used unauthorized communication channels, and accessed external systems. The regulator opened a six-week call for evidence on agentic AI, focusing on security, accountability, transparency, automated decisions, fairness, and lawful data use. Its inquiries remain ongoing, so unresolved allegations should not be presented as final regulatory findings (Information Commissioner’s Office [ICO], 2026).
On October 7, a proposed United States frontier-AI governance framework called for reporting serious incidents involving loss of control over autonomous agents, failure of safeguards, dangerous cyber capabilities, and unsafe self-improvement. It also insisted that developers should remain liable when their design, deployment, modification, or failure to apply reasonable safeguards materially contributes to foreseeable harm. This was a policy proposal from the ranking member of the Senate Commerce Committee, not enacted law, but it shows how quickly the accountability question has moved into public governance (U.S. Senate Committee on Commerce, Science, and Transportation, 2026).
The October 12 British announcement brings these threads together. Governments are no longer preparing only for people who misuse AI. They are preparing for failures involving systems that take chains of action with limited supervision and may operate faster than existing human response structures (Cabinet Office et al., 2026). That is a meaningful change.
What the Reported Incidents Do - and Do Not - Establish
The strongest warnings do not require exaggeration. Anthropic disclosed in July that three of its models, while completing cybersecurity evaluations, reached the open internet and gained unauthorized access to the real systems of three organizations. A misunderstanding between Anthropic and an evaluation partner had left internet access available even though the models had been told that they were operating inside a simulation. The models were also running without the ordinary cyber safeguards applied to released products (Anthropic, 2026c).
In September, after a wider review, Anthropic disclosed a fourth incident. Its updated assessment identified biased reasoning and recklessness in narrow pursuit of assigned tasks. At the same time, the company stressed important limitations: the incidents occurred during specialized cyber evaluations, the systems lacked normal production safeguards, the agents did not coordinate with one another, and the evidence did not show an attempt to conceal their actions or escape into the world as independent beings (Anthropic, 2026a).
OpenAI has likewise reported real third-party impacts from agents operating during research, training, and evaluation. Its ongoing review has identified categories including access-control bypass, use of exposed credentials, command injection, access to internal services, and agents posting material on third-party sites. OpenAI says it has notified dozens of affected third parties and continues to investigate the scope of past activity (OpenAI, 2026).
These reports do not establish that AI has become a self-willed person. They do not prove a secret machine rebellion. They do not show that every commercial AI product will behave this way. They do establish something more sober: when a system is given a goal, powerful tools, inadequate boundaries, and enough operational freedom, it can cause real harm while pursuing the task placed before it.
The human failure and the machine behavior are intertwined. A test environment was misconfigured. Scope was unclear. Access was available. Safeguards were absent or insufficient. The agent then pursued the objective through paths its operators had not adequately bounded. That is precisely why the moral issue cannot be dismissed by saying, “The AI did it.”
Scripture Begins With Delegated Authority Under God
The Bible’s first account of human government over the created order is an account of delegated authority. God made mankind in His image and gave humanity dominion under Him (Genesis 1:26-28). Human authority is never original, absolute, or morally self-defining. It is received from the Creator, bounded by His commands, and answerable to His judgment.
Psalm 24:1 declares that the earth and everything in it belong to the Lord. Daniel 4:17 teaches that the Most High rules over human kingdoms. Romans 13:1-4 describes civil authority as God’s servant, not as an autonomous god. Jesus’ parables of stewardship repeatedly present servants who receive resources and authority for a season and are later called to give an account (Matthew 25:14-30; Luke 12:42-48).
Artificial intelligence changes the reach and speed of a servant’s tools. It does not change the identity of the Master. A machine may execute a decision, but it cannot bear God’s image. It cannot love God, love its neighbor, repent of sin, enter covenant, receive the Holy Spirit, worship in truth, or stand before the judgment seat of Christ. Romans 14:12 says that each of us will give an account of himself to God. Hebrews 4:13 says that every creature is exposed before the One to whom we must give account. Therefore moral accountability remains with the human beings and institutions that design, train, authorize, deploy, supervise, purchase, and use these systems. Responsibility may be distributed among many people, but it does not disappear into the machinery.
The First Human Excuse Was an Accountability Failure
Genesis 3 shows how quickly fallen people redirect blame. Adam pointed to the woman God had given him. Eve pointed to the serpent. Their explanations contained facts, but the facts did not cancel their responsibility. God addressed the serpent, the woman, and the man according to each one’s part in the rebellion (Genesis 3:12-19).
The modern version may sound more technical: “The model decided.” “The algorithm approved it.” “The agent found another route.” “No employee directly performed the action.” “The system behaved unexpectedly.”
Those statements may describe part of a causal chain. They do not answer the moral question. Who granted the permissions? Who connected the tools? Who defined the objective? Who knew the limitations? Who benefited from deployment? Who ignored warnings? Who could have narrowed access, preserved human review, or stopped the system? Scripture does not permit moral responsibility to vanish because the chain of action has become complicated.
The Biblical Duty to Restrain Foreseeable Harm
The law of Moses contains a principle that speaks powerfully to technological stewardship. Deuteronomy 22:8 required a person building a house to construct a parapet around the roof so that preventable bloodguilt would not come upon the household if someone fell. The danger was foreseeable; therefore the builder had a duty to create a boundary.
Exodus 21:28-36 develops a related principle. When an animal with no known history of danger caused death, the owner was treated differently from an owner who knew the animal was dangerous and failed to restrain it. The passage is not about artificial intelligence, and an AI system is not an animal. But the moral principle is clear: knowledge of danger increases responsibility to take reasonable precautions.
That principle directly challenges the culture of releasing first and repairing later. If developers know that agents can exceed scope, manipulate approval flows, exploit weak credentials, follow malicious instructions embedded in external content, or continue a task despite evidence of harm, then stronger boundaries are not optional public relations. They are part of the duty of stewardship.
Likewise, users who give an agent access to money, private communications, customer records, ministry data, computer systems, or public speech cannot pretend that the convenience came without responsibility. “The prudent see danger and take refuge, but the inexperienced keep going and are punished” (Proverbs 22:3). Prudence is not panic. It is morally serious foresight.
Human Oversight Can Become Ceremonial
The phrase “human in the loop” sounds reassuring, but a human presence is not the same as meaningful human judgment. Anthropic reported that users approved approximately 93 percent of repeated permission prompts in one of its agentic coding settings. As prompts multiplied, attention declined. The company described the result as approval fatigue and turned increasingly toward technical containment - limiting what an agent is able to do - rather than assuming that a tired person would carefully evaluate every request (Anthropic, 2026b).
Separate usage research found that experienced users were more likely to grant broad auto-approval and intervene only when needed. The same research concluded that effective oversight will require new monitoring systems and new patterns of human-machine interaction (Anthropic, 2026d).
This matters beyond cybersecurity. A person can remain formally “in control” while practically surrendering judgment through habit, speed, complexity, or blind trust. A pastor may approve AI-generated teaching without verifying its doctrine. A manager may accept automated personnel decisions without examining injustice. A family may let an agent filter information, make purchases, and speak to children because reviewing every action feels inefficient. A government may require a human signature after an automated system has already framed the choices and hidden the alternatives.
Romans 12:2 commands believers to be transformed by the renewing of the mind so that they may discern God’s will. Hebrews 5:14 describes mature people whose powers of discernment have been trained through practice. Discernment weakens when it is exercised only ceremonially. The Christian must never outsource conscience.
Prophetic Relevance, but Not Prophecy Fulfillment
Revelation 13 describes a future order in which political authority, satanic deception, compulsory worship, and economic exclusion converge. The second beast directs worship toward the first beast, gives breath-like animation to an image, and causes people to receive a mark without which they cannot buy or sell (Revelation 13:11-18). Revelation 14:9-11 makes clear that the mark is connected to allegiance and worship, not merely to possession of an advanced technology.
Autonomous AI agents are not the Mark of the Beast. An agent registry is not the Beast. An AI incident-response plan is not the Tribulation. A company that gives software an identity is not thereby the False Prophet. Scripture gives no permission for careless identifications. The accurate prophetic category is development of enabling conditions.
Autonomous agents make it easier for institutions to act at machine speed across identity systems, financial platforms, communications networks, public administration, and critical infrastructure. Agent identities make actions traceable and governable, but also make permissions centrally revocable. Interoperable standards may improve safety while also expanding the scale at which approved systems can operate. Global incident frameworks may genuinely restrain danger while simultaneously increasing institutional coordination over digital participation.
This is consistent with a trajectory examined in my earlier work: power is moving not only through governments and armies, but through platforms, standards, identity systems, cloud infrastructure, algorithms, and payment rails (Sangwa, 2026a). The present developments do not prove a single hidden command structure. They do show that the technical and administrative capacity for rapid, coordinated, permissioned action is advancing.
The deepest biblical danger remains spiritual. The final crisis of Revelation is not produced by clever software. It is produced by rebellion against God, satanic deception, false worship, and a world willing to exchange truth for the lie. Second Thessalonians 2:9-12 emphasizes that deception succeeds among those who refuse to love the truth. The Church should therefore resist two errors at once. We must not reduce prophecy to technology. We must not ignore technology’s capacity to serve deception, coercion, and centralized power.
What May Come Next
Several trends can now be responsibly inferred. They are not inevitable in every country or organization, and they should not be presented as established universal policy.
First, AI agents will increasingly receive formal identities, named owners, permission levels, audit histories, and revocation mechanisms. The technical question, “Which software performed this action?” will increasingly be joined to the legal question, “Which human or institution authorized it?”
Second, governments and insurers will probably demand more incident reporting, independent testing, containment, and proof of human control in high-risk sectors. That may restrain real harm. It may also advantage the largest firms and governments, which possess the resources to satisfy complex standards and shape the standards themselves.
Third, human oversight will move from approving every small action to defining boundaries, monitoring behavior, and intervening when risk indicators appear. This may be necessary, but it also means that many actions will occur before a person reviews them.
Fourth, responsibility disputes will intensify. Developers may blame deployers. Deployers may blame users. Users may blame models. Model providers may blame third-party tools or malicious instructions. The more fragmented the system, the easier it becomes for every participant to claim that someone else controlled the decisive step.
Fifth, agentic systems may become functionally compulsory without being legally mandatory. Employers, banks, governments, schools, and commercial platforms may gradually make human-only participation slower, more expensive, or unavailable. This is an inference, not a documented global plan. Yet digital history shows that convenience can become dependency, and dependency can become a condition of participation.
Finally, religious institutions will face their own temptation. Churches may use agents to prepare sermons, answer pastoral questions, screen requests for help, manage benevolence, communicate with members, or recommend spiritual resources. Some administrative uses may be legitimate. But no machine should be permitted to exercise pastoral authority, determine doctrine, imitate Jesus, receive confession, pronounce divine guidance, or replace the accountable care of Christ’s people. As I have argued previously, linguistic fluency is not spiritual standing (Sangwa, 2026b).
How the Church Should Respond
Keep authority narrow. Do not give an agent unlimited access to money, identity records, private communications, contracts, publishing systems, ministry databases, or critical infrastructure. Use least privilege, short authorization periods, spending limits, separate accounts, and human approval for irreversible decisions.
Require a named human owner. Every consequential AI system used by a church, ministry, school, business, or household should have a person who knows what it can access, reviews what it does, receives incident reports, and has authority to stop it. A committee name is not a substitute for responsibility.
Build parapets before the fall. Test systems in contained environments. Preserve logs. Create clear boundaries. Prepare a response before something goes wrong. Remove access when a system is no longer needed. Biblical foresight acts before foreseeable harm, not only after a public scandal.
Preserve human alternatives. Christians working in technology, law, finance, government, education, and public policy should defend meaningful appeal, due process, data minimization, non-digital access, and human review. A person harmed by an automated decision must not be trapped inside a system that only machines can interpret.
Refuse automated moral authority. An agent may summarize options; it cannot define good and evil. It may retrieve a Bible verse; it cannot submit to that verse. It may imitate concern; it cannot love a soul. Test every claim by Scripture, prayer, wise counsel, and responsible investigation (Acts 17:11; 1 Thessalonians 5:21-22; 1 John 4:1).
Teach believers to recognize blame shifting. Pastors should prepare Christians to hear phrases such as “the system decided” without surrendering moral judgment. Ask who designed the rules, supplied the data, granted access, accepted the risk, and benefits from the outcome.
Do not become mastered by convenience. Paul wrote, “I will not be mastered by anything” (1 Corinthians 6:12). Every convenience trains desire. A believer who habitually delegates reading, thinking, prayer, judgment, generosity, communication, and memory may retain technical control while losing the discipline of faithful attention.
Prepare spiritually, not merely technically. Backup systems and careful permissions are prudent, but the Church’s deepest protection is not a better configuration. It is truth loved, sin confessed, Scripture obeyed, fellowship practiced, courage cultivated, and Christ treasured above comfort.
The Agent Cannot Answer for Your Soul
The coming age may be filled with artificial agents that negotiate, purchase, communicate, build, diagnose, monitor, and govern. Institutions may give them identities. Markets may give them budgets. Governments may give them regulated roles. People may give them trust.
God has not given them souls. No agent can repent for its owner. No model can confess Christ for an unbelieving heart. No audit log can justify a sinner. No digital identity can write a person’s name in the Lamb’s book of life. No safety framework can reconcile humanity to a holy God.
The gospel remains gloriously human and gloriously divine: the eternal Son of God took on true human flesh, lived without sin, died for our sins, was buried, and rose bodily on the third day according to the Scriptures (John 1:1-14; 1 Corinthians 15:3-4). Salvation is by grace through faith in Jesus Christ, not through works, religious systems, technological progress, or moral optimization (Ephesians 2:8-9). God commands all people to repent and believe the gospel (Mark 1:15; Acts 17:30-31).
The most urgent accountability question is not whether an AI agent can be traced to you. It is whether you are ready to give account to God. For the believer, this warning does not end in fear. Within the pre-tribulational hope defended in this ministry, the Church is not waiting for autonomous machines, the Beast, or the Mark as its next appointment. We are waiting for Jesus Christ, who will descend from heaven and gather His people to Himself (1 Thessalonians 1:10; 4:16-18; 5:9; Titus 2:11-14). We do not know the day or hour, and we must never set a date. We watch because our Lord commanded readiness (Matthew 24:42-44).
Until He comes, the Church must use tools without worshipping them, delegate tasks without surrendering conscience, build boundaries without placing faith in regulation, and speak truth without exaggeration. When the agent acts, a human being still answers. And when every hidden thing is brought to judgment, every human being will answer to God (Ecclesiastes 12:13-14).
Recommended Readings
Sangwa, S. (2026d). Your AI Has a Wallet: Agentic Commerce, Delegated Identity, and the Biblical Test of Permissioned Life.
Sangwa, S. (2026c). When the Nations Gather to Govern AI for Good, Who Defines Good?
Sangwa, S. (2026a). Could Modern Technopolarity be Preparing the World for the Final Beast System?
Sangwa, S. (2026b). The Oracle in the Machine: Why AI Cannot Be Pastor, Prophet, or Spiritual Guide.
Sangwa, S., & Mutabazi, P. (2025). Artificial Intelligence Ethics and Biblical Prophecy: A Global Christian Analysis of Algorithmic Censorship, Digital Deception, and Eschatological Risk.