When Can AI Cause Harm
When can AI cause harm? ==AI can cause harm when it is used at scale without appropriate safeguards, oversight, testing, or regard for the context in which it operates.== [:cite[1]{ln=1}] [:cite[1]{ln=2}] It can c...
When can AI cause harm? ==AI can cause harm when it is used at scale without appropriate safeguards, oversight, testing, or regard for the context in which it operates.== [:cite[1]{ln=1}] [:cite[1]{ln=2}] It can cause harm when: It discriminates. Poorly designed systems may produce unfair outcomes for particular groups, especially when their data or design reflects existing inequalities. [:cite[2]{ln=1}] It violates privacy. AI can collect, infer, or expose sensitive personal information; facial recognition and other monitoring tools can enable intrusive surveillance. [:cite[2]{ln=1}] [:cite[4]{ln=5}] [:cite[3]{ln=1}] It strengthens repression. Governments may use AI for mass surveillance, monitoring opposition or minorities, and manipulating public discussion. [:cite[3]{ln=1}] It enables cyberattacks. AI can help attackers conduct phishing, breach systems, and spread deepfakes quickly—especially where cybersecurity capacity and enforcement are weak. [:cite[1]{ln=3}] [:cite[5]{ln=3}] [:cite[5]{ln=4}] It spreads misinformation. AI systems can generate persuasive messages, target persuadable audiences, and optimize misleading content for engagement. [:cite[6]{ln=1}] [:cite[6]{ln=3}] It makes confident but wrong decisions. When a model is developed for one setting and deployed in another, a hidden “contextual mismatch” can produce outputs that appear expert but are unreliable. [:cite[7]{ln=3}] [:cite[7]{ln=4}] It affects people’s rights without adequate review. The risks are especially serious when AI decisions influence public service access, benefits, or other entitlements. [:cite[8]{ln=1}] It disrupts jobs and increases inequality. AI can displace some tasks or jobs, while the gains may flow disproportionately to firms and people who already possess capital, skills, and access. [:cite[10]{ln=1}] [:cite[9]{ln=5}] It damages trust. A visible failure in a high consequence area can erode public confidence faster than successful deployments build it; widespread harm can make future beneficial adoption harder. [:cite[11]{ln=1}] [:cite[11]{ln=2}] ==The safest approach is to test AI with real world data, assess its effects on people, monitor performance after deployment, and verify that it achieves its intended benefits without causing harm.== [:cite[12]{ln=2}] [:cite[12]{ln=3}]