The Dark Side of Artificial Intelligence
by Asma Benmeziane
“There is a 10% to 20% chance that AI will lead to human extinction within the next three
decades”, warned computer scientist Geoffrey Hinton, often touted as “the Godfather of AI”,
during an interview on BBC Radio 4’s Today Programme (qtd. in Milmo). Dr Hinton, who was
recently awarded the Nobel Prize for his foundational work in AI, famously resigned from his
job at Google after working for more than a decade and becoming one of the most respected
voices in the field, to speak more openly about the existential risks posed by uninhibited AI
development, even admitting that a part of him regrets his life’s work. He further emphasised the
urgency of the issue by predicting that “within probably the next 20 years, we are going to
develop AIs that are smarter than people, and that’s a very scary thought”(qtd .in Milmo). This
absence of regulatory boundaries has already allowed AI to move from a theoretical threat to a
present crisis. Ultimately, unconstrained Artificial Intelligence poses a profound threat to society
through unlimited military adoption, frictionless scamming operations, and deteriorating teenage
mental health, making it vital for all of us to understand and respond.
To begin with, AI has transitioned from a supporting tool to a decisive factor in life-or-death
military operations, fundamentally changing the nature of combat. On the battlefield, armies
usually follow a familiar pattern to secure victory, known as the OODA loop (Observe by
predicting the opponent’s moves, Orient through constant army movement suggestions, Decide
by taking private action, and Act by applying the chosen plan) (“Speeding up”). The military’s
push to speed up this process is evident in corporate policy shifts; when Anthropic refused to
remove safety guardrails regarding the military use of its AI, it was effectively blacklisted from
federal contracts, leading the Pentagon to partner with OpenAI instead (Amodei, “Where Things
Stand”). Separately, advanced military targeting systems like the “Gospel” and “Lavender”
generate lists of potential targets (more than a hundred per day) that were previously impossible
for human analysts to process (Davies). Switching from human speed that takes weeks to
software speed that automates data analysis in seconds allows military forces to complete the
OODA loop significantly faster. Consequently, strategists, technologists, and warfighters agree
that AI is a tactical and strategic way to outpace an adversary's decision-making processes,
regardless of shared military equipment, human resources, or combat expertise (“Speeding up”).
Additionally, a study led by Professor Kenneth Payne of the Department of Defence Studies at
King’s College London found a potential Near-Universal Nuclear Escalation Risk ("King's
Study"). Three leading AI models (GPT 5.2, Claude Sonnet 4 and Gemini Flash 3) were placed
in simulated conflict scenarios. After analysing these games, 95% of them saw tactical nuclear
use, and 76% reached strategic nuclear threats (Payne). The reason behind these mind-blowing
findings is that the software is completely emotionless, treating these bombs as strategic pieces
on the board, just like chess, rather than existential threats to humankind, especially Gemini.Only human beings know the tragedy of Hiroshima on August 6th,1945 and Nagasaki on August
9th, 1945. A piece of code that obviously lacks humanity will never understand the cost behind
playing with nuclear power. Contrasting AI's cold logic with human history highlights the
irreplaceable role of moral caution in high-stakes military decisions. During the 1962 Cuban
Missile Crisis, Soviet naval officer Vasili Arkhipov single-handedly prevented a nuclear strike by
refusing to authorise a nuclear torpedo launch despite intense pressure while surrounded by U.S.
forces (Borger). Decades later, in 1983, Soviet officer Stanislav Petrov exercised critical human
judgment when early-warning satellite radars falsely indicated an incoming U.S. missile attack;
recognising it as a technical glitch, he chose not to report it as a strike, averting accidental
nuclear war. Both men saved the world only by exercising human judgment and moral caution
(Borger). This raises a critical moral responsibility for policymakers and military leaders: when
AI makes lethal errors such as targeting civilians, who is held accountable? We must
acknowledge the vital role of human oversight to prevent disaster.
Moreover, AI's rapid yet unpredictable advancement is outpacing current regulations, creating
new avenues for sophisticated scams and financial fraud. A critical examination of voice cloning
technology reveals that it was originally confined to film studios for voice-overs and research
laboratories, but is now accessible to anyone with an internet connection, including criminals.
Modern cloning systems are trained on audio samples to learn the unique acoustic characteristics
of an individual’s voice, including pitch, cadence, accent, tone, and even speech patterns. Once
trained, these models can generate new speech in that voice from any text input through
text-to-speech synthesis, producing audio output that is indistinguishable from the genuine voice
(Adaptive Team). With this immense potential, criminals exploit it to trick victims, fostering fear
and urgency that trap people into financial scams. Raising awareness can help the audience stay
vigilant against these manipulative tactics, protecting themselves and their loved ones from
deception.
One well-known case is Jennifer DeStefano’s virtual kidnapping story. She is an Arizona mother
who suddenly received a call from fake kidnappers using AI to mimic her voice. They claimed
to have kidnapped her and asked for a ransom of $ 1m, saying, “I’ve got your daughter,” and
playing a generated audio clip of her crying and sobbing in the background: “Mom, I messed
up!” The man on the line kept threatening the mother. After many negotiations, they lowered it
to $50,000, only to find out her daughter was in the comfort of her room and had no idea what
happened (Milman). Another, more complex case dubbed “the family emergency scam” involved
scammers targeting lawyer Gary Schildhorn. He received a call from someone whom he believed
was his son Brett, saying that he needed money to post bail following a car crash. He told his
father that he had wrecked his vehicle and needed $90,000 to post bail, that he had a broken
nose, and that he had hit a pregnant woman’s car and instructed his father to call the public
attorney assigned to his case. Later, he discovered that Brett was safe and that everything was a
scam after losing $9,000 (Van Dyk). This scamming process takes advantage of human trust and
manipulates psychology by creating a sense of extreme urgency and fear. Because the voice
sounds incredibly similar to a loved one, victims act immediately without proper verification,
which is exactly what scammers count on.On the other hand, deepfake media is simply a further advancement of Voice Cloning. Modern
deepfake systems are trained on extensive visual and acoustic data to learn a subject's unique
micro-expressions, facial structure, skin tone, lighting interactions, and vocal nuances. Once
trained, these models can swap faces in real-time, generate entirely new video clips, or align lip
movements to foreign languages, producing hyper-realistic media that is increasingly difficult to
distinguish from genuine footage. This brings us to the misuse of this capability, where
high-profile figures like MrBeast have had their faces and voices used in ads to lure followers
into scams, such as fake $2 iPhone offers. These scams are highly effective because they
leverage the public’s trust in recognisable faces as evidence (“Tom Hanks Warns”). This
highlights the massive gap between technological innovation and legal enforcement. Scammers
are highly motivated to use Artificial Intelligence tools because they are scalable, anonymous,
and difficult to prosecute once funds are moved to untraceable digital currencies.
Furthermore, AI technology offers impressive capabilities for interaction, which has led to
forming romantic or deep emotional attachments towards chatbots, which is fundamentally
detrimental to human well-being. Nowadays, the vast majority of people are using ChatGPT and
other modalities to open up about their problems and daily lives. This has happened because of
advances in Natural Language Processing. NLP is a branch of artificial intelligence that enables
computers to understand, interpret, and generate human language. By combining computational
linguistics with machine learning, NLP bridges the gap between complex machine code and
human communication, leading people to treat it as a real human being and slowly giving up on
their human connections. In 1966, MIT professor Joseph Weizenbaum developed a chatbot
named ELIZA to study natural language communication between humans and machines. It took
the position of a text-based therapist; it would ask, “Is something troubling you?” then identify a
keyword in the user’s response, “I’m feeling bad,” and repeat it back as a question, such as “Why
are you feeling bad?” Professor Joseph noticed that this simple question-and-answer series had a
huge effect; he says, “ I had not realised that extremely short exposures to a relatively simple
computer program could induce powerful delusional thinking in quite normal people.”
(Weizenbaum 36). The Eliza Effect, named after this chatbot, describes the human tendency to
project consciousness, empathy, and emotions onto basic computer programs. Even simple AI
can trigger deep psychological illusions of human connections. Based on this, applications like
Replika and Character AI were developed using Machine Learning and Data Scraping
technologies. These apps are promoting adult content and making our teenagers or even adults
attached, isolated and mentally destroyed. Daily interactions with AI have gone from mental
health support to causing depression or death on a societal level. According to a study conducted
by Dr David Levy, an AI researcher and author of Sex and Love With Robots, he predicted that
marriages between humans and AI will be legal by 2050 (Christensen). Because AI is faking
emotions while responding to prompts, even therapists, psychologists, and psychiatrists highlight
the danger of its advice, questioning how it can advise on something it could never experience.
To conclude, while Artificial Intelligence was built to mimic and assist human thought, its
unrestrained expansion now poses an existential threat to the very peace of human life. Aswarning voices like Geoffrey Hinton caution, the lack of regulatory boundaries has allowed AI to
rapidly transition from a speculative concern into an active crisis, fueling autonomous warfare,
enabling sophisticated scams, and destabilising youth mental health. If humanity fails to
establish firm guardrails and governance today, the technology designed to advance our society
may instead surpass our control, making the protection of human existence the defining
challenge of the coming decades.
The Works Cited Page:
Milmo, Dan. "'Godfather of AI' Shortens Odds of the Technology Wiping Out Humanity Over
Next 30 Years." The Guardian, 27 Dec. 2024,
https://www.theguardian.com/technology/2024/dec/27/godfather-of-ai-raises-odds-of-the-technol
ogy-wiping-out-humanity-over-next-30-years.
"Speeding Up the OODA Loop With AI." Joint Air Power Competence Centre, 5 May 2022,
https://www.japcc.org/essays/speeding-up-the-ooda-loop-with-ai/.
OpenAI. "Our Agreement with the Department of War." OpenAI, 2 Mar. 2026,
https://openai.com/index/our-agreement-with-the-department-of-war/. Amodei, Dario. "Where
Things Stand with the Department of War." Anthropic, 5 Mar. 2026,
https://www.anthropic.com/news/where-stand-department-war.
Davies, Harry, et al. "' The Gospel': How Israel Uses AI to Select Bombing Targets in Gaza." The
Guardian, 1 Dec. 2023,
https://www.theguardian.com/world/2023/dec/01/the-gospel-how-israel-uses-ai-to-select-bombin
g-targets.
King’s College London. "King’s Study Finds AI Chose Nuclear Signalling in 95% of Simulated
Crises." King’ s College London, 3 Mar. 2026,
https://www.kcl.ac.uk/news/artificial-intelligence-under-nuclear-pressure-first-large-scale-kings-
study-reveals-how-ai-models-reason-and-escalate-under-crisis
.Payne, Kenneth. "AI Arms and Influence: Frontier Models Exhibit Sophisticated Reasoning in
Simulated Nuclear Crises." arXiv, 16 Feb. 2026,
https://arxiv.org/pdf/2602.1474
Adaptive Team. "The Ultimate Guide to AI Voice Cloning Scams: How to Detect, Prevent, and
Protect Against Them." Adaptive Security, 22 Apr. 2026,
www.adaptivesecurity.com/blog/the-ultimate-guide-to-ai-voice-cloning-scams-how-to-detect-pre
vent-and-protect-against-them.
0Milman, Oliver. "'It Was Entirely His Voice': AI Voice Scams Target Parents in US." The
Guardian, 14 June 2023,
www.theguardian.com/us-news/2023/jun/14/ai-kidnapping-scam-senate-hearing-jennifer-destefa
no.
van Dyk, Deirdre. "Grandparent Scams Take Advantage of Your Love for Your Family." AARP,
13 Feb. 2025, www.aarp.org/money/scams-fraud/grandparent/.
Weizenbaum, Joseph. "ELIZA A Computer Program for the Study of Natural Language
Communication Between Man and Machine." Communications of the ACM, vol. 9, no. 1, 1966,
pp. 36-45. University at Buffalo,
www.cse.buffalo.edu/~rapaport/572/S02/weizenbaum.eliza.1966.pdf.
"Tom Hanks Warns Fans of Dental Plan Ad Using AI Version of Him." The Washington Post, 3
Oct. 2023, www.washingtonpost.com/entertainment/2023/10/03/tom-hanks-ai-ad-deepfake/.
Levy, David. Love and Sex with Robots: The Evolution of Human-Robot Relationships.
HarperCollins, 2007. Google Books, books.google.com/books?id=0061359750.