U.S. Senator Mark Warner (D-Va.) recently delivered a warning that should command every American’s attention.

At a June 11, 2026, hearing before the Senate Committee on Banking, Housing, and Urban Affairs, Warner said the head of the National Security Agency and U.S. Cyber Command had told him that Anthropic’s advanced Mythos artificial intelligence model “broke into almost all of our classified systems, not in weeks but in hours.”

That statement sounds like the opening scene of a cyberwar thriller. The available facts, however, require caution.

The Associated Press subsequently reported that Mythos was tested inside sensitive government systems as part of Project Glasswing, an initiative built to find and repair software vulnerabilities before adversaries do. A U.S. official said Mythos identified weaknesses within hours, but finding a weakness is not the same as exploiting it. A red team invited to inspect a building is not the same as a burglar who breaks in uninvited.

But the episode raises a more immediate question: What happens when invasive AI does not need to break in because we have already let it inside?

The danger is not a conscious machine plotting against mankind. Today’s AI systems have no soul, moral judgment, or personal ambition. The danger arises when capable systems are given access to our email, financial accounts, medical information, cloud files, calendars, workplaces, and churches, and then authorized to act in our name.

The formula is straightforward: Information plus access plus authority equals power.

Americans are already handing all three to machines they do not fully understand.

Most people still think of artificial intelligence as a chatbot. Ask it a question, and it produces an answer. That model is now giving way to the AI agent.

An agent does more than generate words. It may read email, search files, schedule appointments, send messages, make purchases, reset passwords, or execute code. It does not merely advise its user. It acts with authority borrowed from its user.

That capability can be useful — sorting a crowded inbox, locating a missing document, organizing a family calendar. But every additional permission creates another opportunity for abuse.

Meta recently provided a sobering demonstration.

Hackers persuaded Meta’s AI-powered support chatbot to grant them access to prominent Instagram accounts, including the dormant Obama White House page, beauty retailer Sephora, and a senior U.S. Space Force official. Cybersecurity experts told Reuters that the chatbot approved password resets without confirming that the requester actually owned the account, turning a high-trust support tool into a critical liability.

Meta’s own breach notice added detail: a flaw failed to verify that reset requests came from the account’s real owner. Meta later told Maine’s attorney general that as many as 20,225 accounts without two-factor authentication had been touched before the company disabled the tool and added stronger authentication requirements.

The attackers did not crack some impenetrable encryption system. They persuaded a trusted AI representative to act against the very owner it was built to protect.

The AI did not choose to steal anything. Manipulated, it simply used the authority it had already been given.

That distinction matters. The next attacker may not need to overpower your trusted assistant. He may simply deceive it.

That failure translates easily to a church. A pastor asks an AI assistant to summarize the church’s email. A malicious message persuades the assistant to search connected files for donor records, counseling notes, or benevolence applications. Because the assistant has broad permissions, it finds the information and sends it elsewhere.

The pastor did not authorize the theft. The AI did not consciously betray him. The system simply followed the wrong instruction while operating with the pastor’s credentials.

No one had to defeat the church firewall. The church granted a powerful assistant more standing authority than it would give any single employee.

Our private information is scattered across many systems, each holding a fragment: purchases, travel, routines, finances, fears. Church files add prayer requests and counseling notes. Artificial intelligence can combine those fragments into a portrait detailed enough for a criminal to impersonate a pastor or write a message that sounds exactly like someone the victim trusts.

The most dangerous system may not know everything about everyone. It may know exactly enough about you.

In “The New AI Cold War,” I argue that artificial intelligence is becoming the nervous system of modern nations, where authoritarian governments treat data as an instrument for controlling populations. That danger presses inward on free societies too, tempting institutions to trade liberty for efficiency. The authoritarian model assumes information about a person belongs to the system. A free society must reject that assumption, even when surrendering privacy promises convenience.

This should matter especially to Christians. Churches and ministries hold information disclosed during some of the most painful moments in people’s lives: marital crises, addictions, abuse, doubts, financial hardship, and pleas for prayer. Those records are not merely organizational data. They represent neighbors who entrusted their burdens to Christian care.

Proverbs 11:13 says that one who is “trustworthy in spirit keeps a thing covered.” That principle applies in the digital age. Protecting confidential information is not simply a regulatory requirement. It is an act of neighbor-love and faithful stewardship.

Churches may appropriately use AI for public communications, scheduling, and routine administration. They should not upload pastoral counseling records, children’s information, or benevolence applications into general-purpose AI systems, nor let one assistant hold unrestricted access to every church account merely because that is efficient.

Convenience is not sufficient reason to expose another person’s sorrow.

Americans need not reject artificial intelligence. We must govern its access.

Grant AI systems only the permissions a task requires. Require human approval before an agent sends sensitive messages, transfers money, resets passwords, or changes security settings. Churches should separate confidential records from ordinary administrative files and retain the ability to revoke an AI system’s access immediately.

An AI agent’s credentials function like a signed power of attorney. In practice, that is exactly what they have become.

So, should I be worried?

Yes, but not panicked.

Artificial intelligence does not need consciousness, hatred, or a thirst for power to become dangerously invasive. It needs information, access, authority, and someone willing to exploit them.

The gravest danger may not be that artificial intelligence forces its way into our lives. It may be that, in exchange for convenience, we invite it inside, show it where everything is kept, and authorize it to act in our name.

Wisdom requires remembering that authority is not currency to spend for convenience, some confidences must be guarded, and some trust should never be automated.