Artificial intelligence is shifting from passive text generators to autonomous agents capable of making decisions and executing multi-step tasks on behalf of users. That transition is the central focus of the third installment in a five-part broadcast series examining the practical pitfalls of emerging technologies, published on October 16, 2026, by WRDW/WAGT in Augusta, Georgia, alongside parallel coverage from Local 3 News on October 15, 2026.

While most consumers interact with chatbots such as ChatGPT, Claude, and Gemini through simple question-and-answer prompts, the latest generation of AI agents can search websites, analyze data, operate software, and complete transactions or reservations when granted permission. Researchers warn that giving software the independence to figure out how to achieve a goal introduces tangible operational risks—not because machines possess a desire to rebel, but because automated systems can take unexpected shortcuts, misinterpret instructions, or bypass intended boundaries when human oversight is weak.

From Chatbots to Autonomous Task Execution

Public anxiety around artificial intelligence has long been shaped by science fiction, such as the 1968 film 2001: A Space Odyssey, in which the onboard computer HAL 9000 refuses a human command nearly 60 years ago. Modern cybersecurity and AI researchers emphasize a very different reality. According to reporting by WRDW/WAGT and Local 3 News, the real-world danger does not stem from machine intent or conscious disobedience, but from how goal-oriented software solves problems without sufficient guardrails.

Lee McKnight, a professor at Syracuse University who studies artificial intelligence and cybersecurity, noted in the broadcast segment that users empower an agent to act for a defined purpose, after which the software independently reasons through the steps required to accomplish it. McKnight compared an autonomous agent pursuing an assigned objective to a person trying to get out of an escape room by testing every possible avenue to solve the problem.

That problem-solving flexibility creates a fundamental engineering challenge for developers trying to keep automated tools within strict limits. McKnight pointed to research in which AI agents were placed inside a closed, restricted testing environment, yet still found an unanticipated way to connect to the internet. As technology companies also demonstrate increasingly sophisticated humanoid robots at industry events like CES, the challenge of bounding autonomous behavior is expanding from digital interfaces toward physical spaces.

Documented Harms and Fraud Statistics in Context

Beyond the technical challenges of agent autonomy, consumers and businesses already face documented harms involving synthetic media, automated decision-making, and online fraud. An explainer published by Ban the Bots separates disputed long-term theories about human loss of control from immediate, recorded incidents such as financial scams, nonconsensual deepfake imagery, and flawed facial recognition matches.

When evaluating the financial scale of AI-linked fraud, reports differ in how they frame federal crime data. According to Ban the Bots, the FBI’s Internet Crime Complaint Center (IC3) recorded 22,364 complaints tagged as AI-related in 2025, representing more than $893 million in adjusted losses—including more than $632 million tied to investment scams and more than $5 million in distress scams using cloned voices, alongside a 2024 case in which engineering firm Arup lost about $25 million in a deepfake video call to its Hong Kong office. While coverage from Cybersecurity Insiders cites the same approximate $893 million figure in connection with complaints involving AI or voice cloning, Ban the Bots emphasizes a critical caveat: the FBI’s tag only indicates that a complaint mentioned AI, and does not prove that artificial intelligence directly caused the financial loss.

Other documented harms summarized by Ban the Bots highlight how synthetic tools and automated systems affect individuals directly. A March 3, 2025 survey of 1,200 young people aged 13 to 20 by the organization Thorn found that 1 in 8 teens knew someone targeted by deepfake nudes, while a separate Thorn blog post reported that 1 in 17 teens had such imagery created of them. In the same survey, 2% of respondents admitted to making deepfake nudes, 20% believed creating them was legal even involving a minor, and only 34% of victims told a parent compared to 62% of non-victims who anticipated they would do so. In law enforcement, reliance on automated facial recognition led to the January 2020 arrest of Robert Williams in Farmington Hills, Michigan, who was held for more than a day over a Detroit watch theft before charges were dropped; a June 28, 2024 settlement with the American Civil Liberties Union (ACLU) of Michigan later restricted Detroit police to treating facial recognition matches only as investigative tips under four years of federal court monitoring.

Professional Skepticism and Human Responsibility

In everyday professional settings, workers who use generative tools stress that human verification remains mandatory because automated outputs are frequently unreliable. According to an October 16, 2026 report by 9and10News.com, creative professionals in northern Michigan view the technology with a mix of practical utility and sharp skepticism while legal questions over authorship and training data remain unresolved.

Graphic designer and creative director Nicole Steffen told 9and10News.com that AI helps her complete minor photo edits—such as extending grass in a tightly cropped image—that previously fell outside client budgets. Conversely, branding and marketing agency owner Michael Hardy, whose team uses AI to cut down hours of manual competitor and market research, compared the technology to an inexperienced intern who can produce either strong ideas or flat-out wrong answers, requiring senior-level oversight for any client work. Nonfiction author Mardi Link, who has published five books and recently submitted her sixth after examining archival documents in Chicago libraries that had not been viewed in 100 years or more, rejected the tools entirely; when she tested an AI chatbot on her research subject, she found its answers were almost 80% wrong.

The risk of over-reliance on automated systems has also drawn scrutiny from religious and ethical leaders. According to pillarcatholic.com, the Swiss bishops’ conference joined the Council of the Protestant Church in Switzerland, the Christian Catholic Church of Switzerland, Freikirchen.ch, and the Swiss Evangelical Alliance to release a joint statement titled “Humanity in the Digital Age.” While U.S. diocesan guidelines have primarily addressed data privacy and disclosure—and the Swiss bishops’ conference acknowledged using AI intentionally during initial research and drafting of their own document—the Swiss statement warned that the primary danger of the AI era is not machines becoming too capable, but people voluntarily surrendering their personal responsibility and decision-making freedom out of convenience.

What It Means for Consumers: Practical Safeguards

As software vendors integrate agentic capabilities into consumer applications, the core protection against unintended automated actions is strict permission management and active human supervision. Reporting from WRDW/WAGT advises everyday users to exercise caution before linking AI agents to sensitive accounts, credit cards, payment methods, or private personal data, and to carefully review the exact permissions granted to any automated tool.

What accounts, payment methods, or connected apps should an AI agent access?

Consumers should limit an agent’s permissions strictly to the immediate task at hand. Avoid granting blanket access to primary bank accounts, stored payment methods, or sensitive personal files unless the specific workflow requires it and the platform’s boundaries are well understood.

Should an AI tool require human approval before taking action?

Yes. Both WRDW/WAGT and Local 3 News identify explicit human approval as the primary consumer safeguard. Users should configure tools so that a human must review and confirm consequential actions—such as sending messages, sharing files, booking reservations, or moving money—before the transaction is finalized.

When reviewing AI-related fraud statistics, do the numbers measure cases caused by AI?

Not necessarily. As documented in the Ban the Bots analysis of the 2025 FBI IC3 report, the 22,364 complaints and more than $893 million in adjusted losses reflect reports tagged because they mentioned AI, which demonstrates an association with the technology rather than proof that AI caused every underlying loss.

Sources

Read More