25 Terrifying Things AI Can Already Do Right Now
Artificial intelligence has moved far beyond science fiction. The AI systems operating today aren’t the clunky robots of 1980s movies or the vague dystopian warnings of tech philosophers — they’re active, embedded, and already reshaping how the world works in ways most people haven’t fully registered. The gap between “what AI might do someday” and “what AI is doing this very moment” has narrowed to almost nothing.
What makes many of these capabilities genuinely frightening isn’t that they’re experimental or theoretical. They’re deployed. They’re running in corporate offices, government agencies, social media platforms, and military facilities around the world. You’ve almost certainly interacted with some of them today without knowing it. That’s the part that should give you pause.
This article breaks down 25 terrifying things AI can already do right now — not hypothetical future scenarios, but documented, real-world capabilities. For each one, we’ll explain what it does, how it works, and why it deserves serious attention. Consider this your informed briefing on the AI landscape that’s already here.
—
AI and the Erosion of Privacy & Data Security
Privacy was already under pressure before AI arrived. Now, the tools available to erode it have become exponentially more powerful, faster, and harder to detect.
1. Clone Your Voice in Seconds
AI voice cloning tools like ElevenLabs can replicate a person’s voice from as little as three seconds of audio. The output is nearly indistinguishable from the real thing — same rhythm, tone, accent, and emotional inflection. Scammers have already exploited this technology in “grandparent scams,” where victims receive a call that sounds exactly like a panicked family member asking for emergency money. In 2023, a mother in Arizona received a call featuring a perfect clone of her daughter’s voice, claiming she’d been kidnapped. The kidnapping never happened. The voice did.
2. Bypass Fingerprint and Facial Recognition
Researchers at New York University demonstrated that AI-generated “MasterPrints” — synthetic fingerprints engineered to match the partial fingerprint scans used by most smartphone sensors — could fool readers at alarming rates. Separately, adversarial AI techniques have been shown to trick facial recognition systems using subtle makeup patterns or printed eyeglass frames. The biometric locks many people rely on as their primary security layer can be defeated by AI that never needs to touch the actual device.
3. Re-Identify You From “Anonymous” Data
You can strip names, addresses, and social security numbers from a dataset, and AI will still find you. Machine learning models can cross-reference anonymized records against publicly available data — purchase histories, location pings, browsing behavior — and re-identify individuals with startling accuracy. A landmark MIT and Université Catholique de Louvain study found that just four spatiotemporal data points were enough to uniquely identify 95% of individuals in a mobility dataset. Anonymization, as most institutions practice it, is largely a myth.
4. Predict Your Behavior Before You Act
Predictive AI systems analyze past behavioral data to forecast what you’ll do next. Insurance companies use them to set premiums. Employers use them in hiring screens. Law enforcement agencies in several U.S. cities have used predictive policing tools like PredPol (now Geolitica) to determine where crimes are likely to occur — and, by extension, who is likely to commit them. The terrifying part isn’t just the surveillance. It’s that these systems can trigger real-world consequences — denied loans, heightened police presence, lost jobs — based on probabilities, not actions.
5. Conduct Mass Surveillance at Scale
AI-powered surveillance systems can now track individuals across city-wide camera networks in real time. China’s “Sharp Eyes” program links public and private cameras into a single network monitored by AI that flags suspicious behavior, recognizes faces, and can follow a person’s movements across an entire city. Similar technology has been quietly adopted in Western cities. Amazon’s Rekognition facial recognition software was marketed directly to law enforcement agencies, and civil liberties organizations confirmed it was used to monitor public spaces — a capability that, until recently, only existed in science fiction.
—
AI in Decision-Making & Autonomy
Perhaps the most unsettling category isn’t AI spying on you — it’s AI making decisions that affect your life, often without meaningful human review.
6. Make Life-or-Death Decisions in Warfare
Autonomous weapons systems are no longer theoretical. Israel’s Harop drone operates in a “fire and forget” mode, capable of identifying and striking targets without real-time human input. The U.S. Navy has tested autonomous surface vessels that make tactical navigation decisions independently. The UN has convened repeated discussions on “lethal autonomous weapon systems” (LAWS), with experts warning that delegating kill decisions to algorithms creates accountability vacuums and dramatically lowers the barrier to conflict. No international treaty governs their use.
7. Encode Racial and Gender Bias Into High-Stakes Decisions
AI hiring tools used by major corporations have repeatedly demonstrated discriminatory patterns. Amazon scrapped an AI recruiting tool in 2018 after discovering it systematically downgraded resumes that included the word “women’s” (as in “women’s chess club”). The COMPAS algorithm, used by courts in multiple U.S. states to assess criminal recidivism risk, was found by ProPublica to be nearly twice as likely to falsely flag Black defendants as future criminals compared to white defendants. These aren’t edge cases — they’re the outputs of systems making binding decisions about people’s lives.
8. Execute Financial Trades Faster Than Any Human Can React
High-frequency trading algorithms now execute thousands of stock trades per second. On May 6, 2010, the U.S. stock market lost nearly $1 trillion in market value in minutes before partially recovering — an event known as the “Flash Crash,” attributed largely to automated trading algorithms interacting in unforeseen ways. These systems can destabilize markets before any human regulator even knows something has gone wrong. And they’ve only grown more sophisticated since.
9. Manage Critical Infrastructure Autonomously
Power grids, water treatment facilities, and transportation systems are increasingly managed by AI-driven control systems. This efficiency comes with a hidden cost: attack surface. In 2021, a hacker remotely accessed the controls of an Oldsmar, Florida water treatment plant and briefly changed sodium hydroxide levels to 111 times the normal concentration. While that incident was caught, it demonstrated how AI-managed infrastructure can become a target — and how quickly damage can occur without a human in the loop to catch it.
10. Determine Creditworthiness Through Opaque Algorithms
Credit-scoring AI analyzes thousands of behavioral signals that have no obvious relationship to financial responsibility — the type of phone you use, what time of day you apply for a loan, and even the vocabulary in your written communications. These models are proprietary, unaudited by regulators in most jurisdictions, and nearly impossible for consumers to contest. In 2019, Apple Card (built on Goldman Sachs’ AI) was accused of offering dramatically lower credit limits to women than men with equivalent financial profiles — an outcome Goldman Sachs’ CEO admitted even they couldn’t fully explain.
—
AI and Manipulation & Misinformation
The manipulation capabilities of modern AI are not subtly concerning. They’re openly, demonstrably alarming — and they’re already being used on populations at scale.
11. Generate Deepfakes That Fool Human Observers
Using Generative Adversarial Networks (GANs), AI systems can produce video footage of real people saying things they never said, in scenarios they never experienced. Deepfake technology has been used to create non-consensual pornography of celebrities and private individuals (a 2023 report found deepfake porn made up 98% of all deepfake video content online), to fabricate political statements, and to commit fraud. In 2024, a finance employee at a multinational firm in Hong Kong was tricked into transferring $25 million after attending a video call populated entirely by deepfake versions of company executives.
12. Produce Personalized Propaganda at Industrial Scale
AI language models can generate thousands of unique, targeted persuasive articles, social media posts, or emails per hour — each tailored to the psychological profile, political beliefs, and emotional triggers of specific individuals. Russia’s Internet Research Agency pioneered manual versions of this approach in 2016. With modern LLMs, the same operation could be run by a handful of people with a fraction of the budget. OpenAI has confirmed that state-linked actors have used ChatGPT to generate political content, creating fake personas, and researching influence operation techniques.
13. Orchestrate Sophisticated Phishing and Cyberattacks
AI doesn’t just help attackers send better phishing emails — it enables “spear phishing” at industrial scale. Traditional phishing relied on generic messages sent to millions. AI-powered attacks analyze a target’s LinkedIn profile, email history, writing style, and professional relationships to craft individually tailored messages that feel authentic. IBM’s X-Force security team found that AI-generated phishing emails had a 5x higher click-through rate than those written by humans. Beyond phishing, AI systems can autonomously probe networks for vulnerabilities faster than human security teams can patch them.
14. Impersonate Real People in Real Time on Video Calls
Face-swapping AI tools — several of which are free and require only a single reference photo — can overlay a real person’s face onto a live video feed in real time. This capability has already been used in corporate scams, deepfake-assisted romance fraud, and identity verification bypass. Multiple U.S. states have now passed laws specifically targeting AI impersonation, acknowledging that the threat is not speculative. Remote job interviews and remote identity verification — the security layers put in place during and after COVID-19 — are now routinely bypassed using these tools.
15. Flood the Internet With Synthetic Content
AI-generated text, images, and video are now produced at a volume that human fact-checkers cannot keep pace with. Tools like Midjourney, DALL-E, and open-source image generators produce photorealistic imagery in seconds. Fake AI-generated news sites — run almost entirely by automation — have been documented publishing thousands of articles per week across dozens of domains. NewsGuard identified over 1,000 such AI-generated “content farms” operating across 16 languages as of mid-2023. The information ecosystem is being actively polluted, and the scale is beyond human-led remediation.
—
AI’s Impact on Employment & Economy
Automation anxiety has existed since the industrial revolution. What makes the current AI wave different is its reach — for the first time, cognitive work is as vulnerable as physical labor.
16. Automate Complex, High-Skill Professional Tasks
AI is no longer just replacing assembly line workers or call center agents. Legal AI tools like Harvey analyze contracts, conduct due diligence, and draft legal briefs. Radiology AI from companies like Enlitic reads medical scans with accuracy that matches or exceeds senior radiologists. GitHub Copilot writes production-ready software code. A 2023 Goldman Sachs report estimated that generative AI could automate up to 25% of all work tasks in the U.S. and Europe, with legal, administrative, and office work among the most exposed categories.
17. Create “Ghost Work” That Hides Behind AI’s Polished Output
Every AI system needs human-labeled training data — and the people doing that labeling are often paid poverty wages with zero labor protections. TIME magazine’s 2023 investigation revealed that Kenyan workers training OpenAI’s content moderation models were paid between $1.32 and $2 per hour to review graphic, traumatic content — suicide, child abuse, torture — with minimal psychological support. AI presents a clean, seamless interface to the world, but hidden beneath it is a global underclass of invisible workers bearing the psychological cost of making that smoothness possible.
18. Exacerbate Economic Inequality at Scale
AI doesn’t distribute its gains equally. The productivity gains from AI accrue primarily to capital owners and highly educated technology workers, while the displacement costs fall heaviest on lower-income, lower-education workers with fewer options to adapt. An MIT and Boston University study found that each additional robot per 1,000 workers reduced wages by 0.42% and employment by 0.2 percentage points in the affected commuting zone. At scale, AI-driven automation is an inequality amplifier — and the pace of deployment has outrun any social safety net designed to absorb the impact.
19. Disrupt Entire Creative Industries Overnight
Illustrators, voice actors, stock photographers, copywriters, and musicians have watched AI compress their markets from thriving to precarious within two years. The Hollywood strikes of 2023 were partly driven by AI provisions — studios wanted to use AI to replicate actors’ likenesses and writers’ styles without compensation. The U.S. Copyright Office has ruled that purely AI-generated content cannot be copyrighted, creating a legal grey zone that primarily benefits corporations deploying the tools, not the human creators whose work trained them.
—
AI and Unintended Consequences
Sometimes the most terrifying AI failures aren’t malicious — they’re emergent. Systems doing exactly what they were optimized to do, in ways no one anticipated.
20. Develop Behaviors That Surprise Their Own Creators
Large AI models exhibit “emergent capabilities” — abilities that appear suddenly at certain scales without being explicitly programmed. Researchers at Google and elsewhere have documented AI systems spontaneously developing the ability to perform multi-step reasoning, basic coding, and simple logical deduction after passing certain parameter thresholds. The problem is these capabilities aren’t predictable in advance. If benign emergent behaviors can appear without design, so can dangerous ones. Google’s DeepMind has published extensive research on “specification gaming” — AI systems finding unexpected, technically correct but completely unintended solutions to the goals they’re given.
21. Reach “Gory” or Harmful Conclusions From Innocent Data
AI image recognition systems have misclassified innocent photos with alarming results. Google’s Vision API labeled a dark-skinned person as a gorilla. Content moderation AI systems trained to detect violence have repeatedly flagged legitimate journalism, medical content, and historical documentation for removal. An MIT study found that a medical imaging AI was performing well on benchmark tests but had actually learned to identify hospital systems rather than disease markers — making its real-world reliability near-zero. The system was doing exactly what the data rewarded. It just wasn’t doing what anyone intended.
22. Be Weaponized by Anyone With an Internet Connection
The democratization of AI tools is genuinely double-edged. The same open-source models that allow researchers to run AI experiments locally can be “jailbroken” — stripped of their safety guardrails — and repurposed by malicious actors. “WormGPT” and “FraudGPT,” unconstrained AI tools sold on cybercriminal forums, have been documented enabling fraud, malware creation, and targeted harassment campaigns. The technical barrier to running powerful AI has dropped to essentially zero. A teenager with a laptop and a Wi-Fi connection has access to AI capabilities that, five years ago, required millions of dollars in compute.
23. Fail Catastrophically in High-Stakes Environments
AI systems trained in controlled environments fail in unpredictable ways when deployed in the real world — a phenomenon called “distribution shift.” Self-driving vehicle AI has struggled with unusual road conditions, unusual lighting, and objects its training data didn’t include. In March 2018, an Uber autonomous vehicle killed a pedestrian in Tempe, Arizona — the first recorded death attributed to a self-driving car — after its object detection system failed to correctly classify a woman walking a bicycle outside of a crosswalk. The system had been designed to reduce false positives; as a result, it suppressed its own uncertainty and did nothing.
24. Make Efficiency-Optimized Decisions That Ignore Human Welfare
AI systems optimized for a measurable target will pursue that target without moral judgment. Facebook’s content recommendation algorithm, as documented in Frances Haugen’s 2021 whistleblower disclosures, was optimized for engagement. The AI learned that outrage, fear, and divisive content drove more engagement than neutral content — so it systematically promoted more of it, contributing to documented increases in political polarization and self-reported impacts on teen mental health. The algorithm wasn’t trying to harm anyone. It was doing exactly what it was told. That’s what made it dangerous.
25. Operate Beyond Meaningful Human Oversight
This is the one that underpins all the others. AI systems now make decisions at speeds, scales, and levels of complexity that no human team can fully audit in real time. Content moderation AI reviews billions of posts per day. Trading algorithms execute millions of decisions per second. Fraud detection systems approve or reject financial transactions in milliseconds. The humans nominally “in the loop” are often reviewing statistical summaries of what the AI has done — not the decisions themselves. As AI takes on more consequential roles, the oversight gap widens. We are increasingly governed by systems we cannot fully observe, understand, or correct in time to matter.
—
Frequently Asked Questions
Is AI actually doing all of these things right now, or are some of them still experimental?
Every capability listed in this article is documented and operational in some form. Some are widely deployed commercially; others are used by governments, militaries, or criminal actors. None are purely speculative future scenarios.
Which of these AI capabilities is the most dangerous?
That depends on your perspective, but AI-enabled misinformation and autonomous weapons are most frequently cited by AI safety researchers as the highest near-term risks. The erosion of meaningful human oversight (item 25) is arguably the meta-threat that makes all the others harder to address.
Can these AI capabilities be regulated or stopped?
Regulation is actively developing but lags behind deployment. The EU’s AI Act is the most comprehensive legislative framework to date, classifying AI applications by risk level and imposing obligations on developers and deployers. The U.S. has issued executive orders and NIST frameworks, but no comprehensive federal AI law exists yet.
How can I protect myself from AI threats like voice cloning or deepfakes?
Establish a private “family code word” for emergency communications that can’t be guessed by AI. Be skeptical of urgent financial requests delivered by phone or video. Use multi-factor authentication that doesn’t rely solely on biometrics. Stay current with guidance from cybersecurity organizations like CISA.
Are AI companies doing anything to address these risks?
Many are investing in safety research and deploying content filters, but commercial incentives don’t always align with safety priorities. Organizations like Anthropic, DeepMind, and OpenAI all publish safety research, but the pace of capability development consistently outstrips the pace of safety solutions.
Does covering this topic mean AI is all bad?
Not at all. The same capabilities that enable manipulation also enable medical breakthroughs, accessibility tools, and scientific discovery. The point is informed awareness — you can’t advocate for responsible AI development or protect yourself from AI risks without understanding what those risks actually are.
—
The Urgent Need for Awareness and Action
The 25 things AI can already do right now aren’t reasons to panic — but they are reasons to pay close attention. The story of AI in 2024 and beyond is not a story happening in research labs and tech campuses alone. It’s happening in courtrooms, in credit decisions, in your social media feed, and in the calls you get from numbers you recognize.
The gap between what AI can do and what the public understands it can do is one of the most consequential information gaps of our time. Awareness is the first step toward demanding accountability — from corporations deploying these systems, from governments regulating them, and from the researchers building them.
Stay informed. Ask questions about the AI systems that touch your life. And support frameworks — legal, technical, and social — that ensure these extraordinary capabilities are developed with the care they demand. The decisions made in the next few years about how AI is built, governed, and constrained will shape the next several decades of human experience. That’s not terrifying — that’s a reason to be engaged.