By Richard W. Stevens

As you and your spouse head off for a two-week Maui vacation, do you hand your 17-year-old boy a bottle of smooth whiskey and the keys to your spare Ferrari, bidding him “have fun while we’re gone”?

Let’s assume sensible parents wouldn’t do that. Yet wizards of tech and captains of biz seem eager to unleash AI autonomous self-actuated (agentic) systems with greater powers of mischief and fewer restraints.

Watching the 2016 Black Mirror episode, “Shut Up and Dance,” after reading more about the newest AI systems, conjured images of digital booze and fast cars. The episode delivers a precise blackmail scheme practically scripted for AI, even before the current era of agentic bots.

Originally sci-fi, the episode describes a horror beyond shock. It comes from structure that AI today smoothly enables. Its entire blackmail scheme operates from text messages.

Cameras Are Watching

Here’s the story. Kenny comes home after work. Opening his laptop, he discovers it’s infected with malware. He installs what he thinks is protection. In fact, the protection hands control to the attackers.

They activate his webcam and record him self-pleasuring to illegal videos. They send proof. The first text message warns Kenny: Comply, or everyone sees the video. That establishes leverage based upon fear and shame. The attackers don’t argue or negotiate. They show power.

An AI agent could handle this stage easily:

• Send phishing links in quantity to lure user downloading.
• Install webcam malware.
• Record video automatically.
• Scan footage for explicit behavior.
• Flag high-leverage victims.

Computer vision systems already detect nudity and sexual activity. A malicious agent could sort victims in seconds. No human review required.

Stage One: Movement, Deadlines, Control

Kenny receives another text telling him to travel 15 miles in 45 minutes to another meeting. No time to think. No time to confess.

At the destination, he meets another text blackmail victim, who hands him a cake box. Next texted instruction: deliver the cake to a specified hotel room. Kenny complies again and meets Hector there. Hector had arranged to meet a prostitute, which the attackers can prove. He fears losing custody of his children. The system has linked the two victims.

More texted instructions come. Deliver the package. Wait together. Drive to new coordinates.

By pairing victims, the attackers increase pressure. Each man now knows the other has something to lose.

An AI agent could coordinate this like a logistics platform:

• Track phone locations.
• Assign meeting points.
• Adjust deadlines.
• Confirm arrivals in real time.

Delivery apps already solve harder routing problems. Blackmail logistics would not strain modern AI.

Stage Two: Escalate the Stakes

The cake Kenny is carrying contains a hidden handgun. A text orders the two men to rob a specified bank. Hector drives. Kenny carries the gun. Kenny enters the bank, points the weapon, takes the cash, and runs. Meanwhile, a separate text message tells Hector to destroy the car.

The plan does more than extract obedience. It deepens exposure. The victims now face felony charges. Even if the original secrets surface, the robbery guarantees legal consequences.

An AI agent could manage this escalation step by step:

• Demand photo proof.
• Monitor timestamps.
• Trigger the next task only after confirmation.
• Escalate threats if compliance slows.

Agentic systems already use conditional logic. If X happens, do Y. If not, increase pressure.

Stage Three: Reveal the Full Truth

Kenny gets another vehicle and drives to a specified location with the stolen money. There he meets Paul, another victim. Paul confesses something darker. The attackers targeted them for viewing child exploitation material. Beyond embarrassment, that is criminal depravity.

The final instruction to Kenny and Paul arrives: The two men must fight to the death. A drone arrives to film them. The winner is promised he can keep the stolen money.

The trap has closed. Utterly despairing, Kenny tries to shoot himself, pulls the trigger, click!  No bullets. Paul seizes the initiative and attacks him.

Meanwhile, Hector has gone home, thinking his cooperation might protect him. Instead, he receives a Trollface image on his phone. His wife already knows.

Kenny apparently won the battle with Paul but returns home to his mother crestfallen and berating him for the videos he watched and the video of his behavior. Police arrest him for his several crimes. The attackers release everything about all of the blackmail victims. Compliance never mattered.

The structure becomes clear:

• Extract obedience.
• Escalate criminality.
• Expose everyone anyway.

An AI agent could automate that entire chain. It could schedule disclosures in advance. It could trigger releases based on timers, not emotions.

No hesitation. No mercy.

How an Autonomous Agent Could Run This

Agentic AI systems already create plans, break tasks into steps, monitor progress, adapt and respond to new inputs, and operate in parallel. A malicious version could:

• Scan the internet for targets.
• Infect devices automatically.
• Analyze webcam footage.
• Build psychological profiles from social media.
• Generate customized threats.
• Coordinate victims geographically.
• Accept cryptocurrency payments.
• Release scandalous material automatically.

The agentic AI system would not need constant human control. A criminal operator could set a high-level goal such as “maximize revenue per victim.” The AI could handle the rest.

Deploying Thousands at Once

The real shift comes with scale. A human blackmailer can handle a handful of victims. An AI cluster could handle thousands. Cloud servers make deployment cheap.

Because AI agents can queue tasks and run them in parallel, they don’t bottleneck. They don’t sleep. They don’t panic.

If law enforcement shuts down one node, the system spins up another. If a payment method fails, the agent switches wallets. If a victim hesitates, the system releases partial proof.

Automation turns blackmail into infrastructure.

Why This Changes the Threat Landscape

The Black Mirror episode unsettles us because the texts are unrelenting. They arrive at the tactically-timed moments and corner the victims at every turn. That kind of precision fits AI perfectly.

AI excels at:

• Real-time monitoring
• Pattern detection
• Deadline enforcement
• Message optimization

A system could test different threat styles across thousands of victims. It could learn which phrasing produces faster compliance. It could adjust automatically. Over time, the workflow improves itself.

From Fiction to Plausible System

In 2016, “Shut Up and Dance” seemed extreme. It felt like the work of cruel, coordinated humans pulling strings behind the curtain. Today, the project reads like a software design:

• Capture leverage.
• Impose deadlines.
• Escalate crimes.
• Interlink victims.
• Automate exposure.

No mastermind required. Just infrastructure. Malign coercion becomes autonomous, scalable, and optimized. The crimes aren’t random or one-up. The project becomes a soulless goal-seeking system. The most disturbing part is not the cruelty. It’s the efficiency.