Is AI dangerous?
AI can cause harm today through errors, bias, scams, deepfakes and privacy loss, and some researchers warn that future, far more capable systems could pose catastrophic risks. Others think those long-term fears are overstated. Most experts agree that safety testing and human oversight matter.
The full answer
AI can be dangerous, but the risks differ in kind and in certainty.
Harms happening now:
- Wrong answers presented confidently (hallucination).
- Bias in systems that screen people for jobs, credit or benefits.
- Fraud and deepfakes, including cloned voices and fake images.
- Privacy loss and misuse of personal data.
- Security risks, such as prompt injection against AI agents.
Risks that are debated: that much more capable systems could be misused for serious harm, or could pursue goals that conflict with ours. Some researchers, including authors of books on our reading list, argue this could be catastrophic. Others argue the danger is overstated and distracts from present harms. The honest summary is that experts disagree.
What lowers the risk: testing before release, human oversight of important decisions, clear rules on who is accountable, and not giving AI agents more access than a task needs. For everyday use, check important facts, do not share private data you would not want stored, and be wary of unsolicited messages that sound too convincing.
Read moreSuperintelligent AI risks in 2026: why deterrence still mattersThe 30 best books on AI and superintelligence