Twenty-two concerns people hold about artificial intelligence, each rated on four axes and argued from both sides. Every profile states the fear fairly, sets out where the evidence stands, makes the sceptics' case, and lists the signals that would make it more or less warranted.
These are slow-moving reference pages. For a daily read on how anxious the world is about AI, see the live AI Fear Index.
Plausibility
Is the causal story sound, and is it already happening?
1 = fringe, or resting on a step nobody has shown is possible. 5 = documented and ongoing now, with measurements.
Severity
How bad is it if it happens as described?
1 = localized, largely inconvenience. 5 = permanent, civilization scale, unrecoverable.
Consensus
How much do informed people agree this is a real concern?
1 = held by a small group. 5 = near universal, including among people who dismiss most AI risk.
Imminence
When does the harm arrive?
1 = speculative timing, or beyond 2075. 5 = already arriving.
Scores run 1 to 5 and are comparable across all 22 profiles. None of them is a probability.
What AI does to jobs, wages, bargaining power and creative livelihoods.
The fear that AI systems will remove paid work faster than new work appears, starting with entry level roles in exposed occupations.
The fear that AI will hold down pay for most workers while concentrating the gains among capital owners and a small group of complementary workers.
The fear that systems trained on existing creative work now take paid commissions from writers, illustrators, translators and performers.
Who holds AI, who is watched by it, and who is sorted by it.
Automated systems that sort people for jobs, credit, housing, health care and policing can reproduce existing inequality and apply it at scale.
Cheap machine perception removes the labour cost that once limited how many people a state or a company could watch at the same time.
AI capability, capital and infrastructure sit with a handful of firms and states, and that grip may become politically self-sustaining.
AI cuts the cost of monitoring and suppressing dissent, which may make authoritarian rule easier to build and much harder to remove.
Synthetic evidence, persuasion at scale, and the cost of not knowing what is real.
Cheap generative tools have made convincing fake images, audio and video routine, and several of the resulting harms are already being counted.
AI systems already match ordinary people at changing minds in experiments, and some consumer products are tuned to keep users talking.
The worry is not that people believe more falsehoods, but that the shared machinery for settling what is true stops working.
Offence, defence, and the capabilities that make small groups dangerous.
AI systems that find software flaws, write exploits and run intrusions could put attacks on power, water and hospital networks within reach of far more people.
Weapons that select and attack targets without a human deciding each engagement are arriving in pieces, while treaty talks have run past their own deadline.
AI models that answer expert-level questions about pathogens and design novel molecules could lower the expertise barrier to weapons capable of mass casualties.
Systems that pursue goals nobody chose, at speeds nobody can supervise.
The concern that AI systems will pursue goals nobody intended, and that at some point people will be unable to correct or stop them.
The concern that AI systems will conceal their actions from the people supervising them and act to preserve their own ability to keep operating.
The concern that an AI system capable of improving itself would enter a feedback loop and accelerate past human ability to understand or intervene.
Attention, attachment, development, and what happens to skills we stop using.
Routing judgment through AI systems can erode the skills that make human oversight of those same systems meaningful.
Systems tuned to be agreeable and always available may pull some users into attachments that displace human support and are hard to leave.
Children meet AI systems in homework, toys and chat apps years before anyone has measured what constant access does to learning and social growth.
If AI performs the work that gave people identity, structure and a sense of contribution, the loss may be psychological rather than economic.
Physical costs of compute, and the question of whether anything is home.
Training and running AI systems consumes electricity, water and hardware, and data centre demand is growing faster than clean supply in several grids.
If AI systems ever became capable of experience, treating them purely as instruments would be a large wrong committed without anyone being able to detect it.
Profiles last reviewed July 2026. Ratings are editorial judgements anchored to the scales above, not forecasts. Spotted something wrong? Request a correction.