Understanding the risks of misusing Artificial Intelligence

The risks of AI 

Artificial intelligence is best understood as a general-purpose technology with built-in inherent risks. The system that helps draft a work email can also help draft a scam. At the moment, public discussion often centres on AI acting on its own, but the more immediate risk is human: what people deliberately do with these tools, and how carelessly they sometimes use them. This article covers three areas: criminal use, military use, and negligent use. 

Criminal Use

Fraud is the clearest example. Language models can write fluent, personalised phishing messages, removing the spelling errors that once exposed many scams. Voice-cloning tools can imitate a person from a short audio sample, and cloned voices have been reported in “family emergency” and payment-authorisation scams. Image and video generators can produce forged documents, fabricated evidence and non-consensual intimate imagery.

AI also lowers the skill barrier. Someone with limited technical ability may get help writing malicious code or planning a social-engineering attack. How much extra capability this gives criminals is still debated. Developers build safeguards into their products, but safeguards are imperfect, and openly released models can be modified to remove them.

Military Use

Armed forces are adopting AI for intelligence analysis, logistics, surveillance, cyber operations and, in some systems, target identification. The concerns are specific:

• Speed: Automated systems can compress the time humans have to deliberate, increasing the risk of escalation by error.
• Accuracy at scale: A misidentification by one soldier is a single mistake. A flawed model can repeat the same mistake thousands of times.
• Accountability: When a system contributes to a harmful decision, responsibility is harder to assign.
• Arms-race pressure: States may deploy systems before they are adequately tested because they fear falling behind rivals.

International discussion of autonomous weapons is ongoing, however there is no binding global agreement governing the use of such weapons.

Negligence and misuse

Not all misuse is intentional. Large language models generate plausible text, not verified facts, and they can state falsehoods with complete confidence. This is often called “hallucination.” In 2023, lawyers in a New York, US federal case were sanctioned after submitting a brief containing case citations that a chatbot had invented. The tool did not fail unusually; the users simply did not check its output and this is by no means an isolated example.

 
Three negligent habits recur:

• Over-trust: Relying on AI output for medical, legal or financial decisions without independent verification.
• Vague instructions: Poorly specified prompts tend to produce answers that look right but miss the actual need, and the user may not notice.
• Excess access: Connecting AI agents to email, files or payment systems without limiting permissions or monitoring actions. A mistake then becomes an action with real consequences.

Psychologists call the underlying tendency 'automation bias': people defer to automated systems even when their own judgement would have caught the error.

Reducing the risk
No single measure removes these risks, but several help:

• Treat AI output as a draft that needs checking, especially where errors are costly.
• Keep a human responsible for consequential decisions.
• Give AI tools the minimum access required for the task.
• Be specific about what you want, and test the result against known facts.
• Support clear rules and transparency from developers, businesses and governments.

Summing up: AI’s risks come largely from how people choose to use it and how carefully they supervise it. Criminals exploit its capabilities, militaries may deploy it faster than they can test it, and ordinary users may trust it more than its reliability justifies. Understanding these patterns is a practical first step toward using the technology responsibly