what sign up
nixm nixm

← learn

Is AI dangerous? Yes, in specific ways. Here they are, ranked.

October 14, 2026

AI is dangerous the way electricity is dangerous: in specific, mostly known ways, and more from how it is used than from what it is. The risks hurting people today are scams with cloned voices, fake images, confidently wrong advice taken as fact, personal data kept and leaked, and biased decisions nobody can appeal. The risks building are to jobs, to concentration of power, and to our own skills. The risk that gets the headlines, an AI that escapes control, is taken seriously by serious people and is not the one that should change what you do this week.

How to think about the question

"Is AI dangerous" bundles three different questions: is it hurting people now, is it building toward harm, and could it ever be catastrophic. They have different answers and different time scales. Ranking them by how much they should change your behaviour this week puts the boring ones first, which is where they belong.

Tier 1: harms happening now

  • Scams with cloned voices and faces. A call from "your daughter" in trouble, a video call from "the CEO" asking for a transfer. Thirty seconds of audio is enough to clone a voice. Do this week: agree a family password for emergencies; treat any urgent money request by voice or video as unverified until you call back on a known number.
  • Fake images and video of real people. From political forgeries to non-consensual sexual images, which have ruined lives, mostly women's and girls'. Laws are catching up unevenly. Do: never treat a generated-looking image as evidence; know that content credentials exist and are starting to appear in real photos and generated ones.
  • Confidently wrong advice taken as fact. Medical, legal, financial. The model answers in the tone of an expert and is sometimes inventing. Do: for anything with consequences, ask for sources and open them, or ask a person who is accountable.
  • Your data, kept. Most consumer AI keeps what you type and many train on it; chats have been exposed in breaches, surfaced in search engines through shared links, and kept under court order. The health question, the contract, the thing about your boss: all of it can outlive the chat. Do: turn off training where you can; use a tool that keeps no transcript for anything you could not retract. The privacy concerns page is the long version.
  • Biased decisions with no appeal. Hiring filters, loan models, benefit systems that learned past unfairness and apply it at scale with nobody to argue with. Do: if a decision about you was automated, you usually have a right to ask for a human review. Use it.

Tier 2: harms building

  • Work. Not mass unemployment this year; the slow removal of routine tasks and the junior rung that trained people. The jobs page ranks what is shrinking.
  • Dependence. If the tool writes the essay, the person did not learn to write the essay. If it answers every question, the habit of working something out withers. This is the danger most under-discussed and most within your control: use it to start, not to finish.
  • Concentration. A few companies decide what billions of people are told, with the economics of training ensuring few can compete. The response is political and slow.
  • Cheaper attacks. Phishing that reads like a native speaker, malware written faster, disinformation at volume. Defence improves too; the balance is unclear.

Tier 3: the catastrophic question

Could an AI system become capable enough and misaligned enough to cause harm no one intended and no one can stop? Serious researchers, including people who built these systems, consider this a real question worth real effort, and the major labs run safety programmes aimed at it. It is also unproven, contested, and a decade-scale concern. The honest position for a non-specialist is: it is right that people work on it, it is not a reason to lose sleep, and it should not crowd out the tier-one list, which is hurting real people today.

What is being done

The EU's AI Act (2024) bans some uses outright, such as social scoring, and puts obligations on high-risk systems. Content credentials are spreading so that real and generated media can be told apart. Labs publish safety testing and have agreed to external evaluation of frontier models. Courts are deciding the training-data cases. None of it is finished; all of it is more than existed three years ago.

So: should you use it?

Yes, with the tier-one habits: verify what matters, assume a voice or video could be fake, and do not hand your private life to a tool that keeps it. That last habit is the one we built nixm around: no transcript, nothing trained on, and a conversation that dissolves after seven days. The danger of AI is mostly in the defaults. Change the defaults and most of it goes away.

frequently asked

Is AI dangerous?

In specific ways, yes. The harms happening now are scams with cloned voices, fake images, confidently wrong advice, personal data kept and leaked, and biased automated decisions. Structural risks to jobs and concentration of power are building. Catastrophic loss-of-control scenarios are taken seriously by researchers but are unproven and long-term.

What is the biggest danger of AI right now?

For an individual, fraud: cloned voices and faces used for urgent money requests. Agree a family password and verify any such request on a known number before acting.

Can AI take over the world?

There is no evidence that current systems could. Some researchers consider future loss-of-control a real long-term risk and work on it; it is not a reason to change what you do this week, unlike the everyday harms.

Is it safe to tell AI personal things?

Only in a tool that does not keep or train on what you type. Most consumer AI stores conversations by default. For anything you could not retract, use a tool that keeps no transcript.

Why is AI dangerous for students?

Mostly dependence: if the tool does the assignment, the student does not learn the skill. Used to explain and check rather than produce, the same tool is a tutor.

What is being done to make AI safe?

The EU AI Act bans some uses and regulates high-risk ones, content credentials help tell real from generated media, labs run safety testing with external evaluation, and courts are settling the training-data questions.

try it now — no account, no transcript kept. even the signal dissolves in seven days.

seed a signal →

Enter your email and receive a link to sign in. No passwords, just a secure signal to connect.

why connect

  • 20 free credits to start
  • your signals, remembered
  • buy more credits anytime
  • image & video generation features
or
+