Fifty-seven percent of Americans say AI is bad for society. That number comes from the ABC News/Washington Post/Ipsos poll published on 21 October 2026, and the intensity sits with the pessimists: 17% call it extremely bad, against 2% who call it extremely good. Sixty-five percent see a high risk of AI eliminating many jobs. Your operators, inspectors, and line supervisors are in that sample. They read the same headlines you do, and most of them have already formed an opinion about your rollout before you announce it.

AI public opinion is now a deployment risk, not a media topic. Below, what the poll data actually tells you about the people on your shop floor, and how to plan an implementation that survives contact with them.

Your AI Rollout Plan Assumes Buy-In That 57% of Americans Don’t Have

Only 42% of Americans told ABC News, the Washington Post and Ipsos that AI is good for society. On the personal question the country splits almost evenly, 50% good for them against 49% bad, but five times as many people say AI is extremely bad for them personally (15%) as say extremely good (3%). That is the mood your inspectors and QA analysts bring to the kickoff meeting.

Most implementation budgets cover licences, integration work, and a two-hour training session. They treat adoption as a skills gap. It is a trust gap, and the two fail differently.

A skills gap produces questions. A trust gap produces quiet compliance: the tool gets opened once a week, the override button gets pressed, and the pilot dies without anyone declaring it dead.

What the Poll Actually Found: Society Bad, Personally Neutral

The headline figure is the least useful part of this poll. The cross-tabs are where the deployment risk sits, and they point in a direction most rollout plans get backwards. One signal to start with: about two-thirds of Americans oppose building new data centers in the U.S., and opposition climbs higher when the question is about their own community. Proximity makes people more hostile, not less.

The society-versus-self gap is where workplace adoption lives

People judge AI harshly in the abstract and far more softly when it touches their own circumstances. The partisan breakdown makes the point sharply: 57% of Republicans say AI is personally good for them, while 55% of Democrats say it is personally bad. Same technology, same week, opposite verdicts. Abstract opinion is tribal. Personal opinion is experiential.

That gap is your opening and your trap. Ipsos found 62% of Americans see a high risk of people losing the ability to make their own decisions without AI, rising to 67% among those under 50. For a quality inspector whose value is judgement, that is not a philosophical worry. Deploy a system that overrides calls instead of informing them, and you confirm the fear on day one.

Younger workers are the most skeptical, not the least

Sixty-five percent of Americans under 30 say AI is bad for society. So do 59% of those aged 30 to 39. Among people 40 and older, it drops to 53%. The generation your change management deck assumes will carry adoption is the generation most convinced the technology is harmful.

Stop appointing your youngest technicians as pilot champions on the assumption that comfort with software equals support for automation. It does not. Pick champions on floor credibility: the inspector other inspectors call when a reading looks wrong, the supervisor who has survived three ERP migrations. Age tells you nothing useful here. Standing does.

The Three Risks Your Workforce Is Actually Worried About

The poll asked Americans to rate specific consequences of AI growth, and three of them land squarely in a plant. Each one shows up as a question you will get asked in a corridor, not in a meeting. Answer them badly and the pilot dies quietly.

Security and data residency: 73% high-risk and the questions it generates

Nearly three-quarters of Americans, 73%, say there is a high risk of AI hacking secure computer systems. That concern crossed every partisan group and every age group in the sample, which makes it the most broadly shared fear in the whole survey. On a shop floor it does not sound like cybersecurity. It sounds like “where do the inspection photos go?”

Your maintenance lead will ask whether supplier records leave the building. Your QA manager will ask whether process data trains someone else’s model. The straight answer names the hosting region, the retention period, and who can query what. If you cannot say which tenant the data sits in, say that, then fix it before the kickoff.

Job elimination and deskilling: the two objections that stop pilots

Sixty-two percent say there is a high risk of people losing the ability to make their own decisions without AI, rising to 67% among Americans under 50. For quality professionals, whose entire value is calibrated judgment, that is the sharper of the two fears. An inspector who suspects the system will make them worse at their job will use it exactly as little as they can get away with.

The answer is not a reassurance. It is a design choice. Keep the human as the decision-maker of record on disposition calls, let the model do detection and triage, and publish the override rate so people can see judgment is still being exercised. When someone asks whether headcount is the real goal, give a number: how many roles, what happens to them, and by when. Vague answers confirm the fear.

What Changes in Practice: Designing Rollouts for Skeptical Teams

Nothing in the poll says your project fails. It says your sequencing and your language matter more than your model choice. Two decisions carry most of the weight: which task you automate first, and how you describe the payback.

Pick first use cases that remove drudgery, not judgment

Start with the work people already complain about. Manual data entry into the QMS, assembling the monthly quality report from six spreadsheets, reconciling supplier certificates against incoming inspection records, chasing down document revisions before an audit. Nobody defends that work in a performance review. Taking it away reads as relief, not replacement.

Do not start with defect dispositioning, root cause analysis, or anything an inspector points to when asked what they are good at. That is where the 62% who see a high risk of people losing the ability to make their own decisions without AI will dig in, and they are not wrong to. Keep the AI as the preparer and the human as the decision-maker: the system drafts the deviation summary, the engineer signs it. Publish where the data goes, which vendor processes it, and who inside the company can read the output, before anyone asks. Answering that question late makes it sound like a cover-up.

How to frame ROI without triggering the job-loss objection

Sixty-five percent of Americans say there is a high risk of AI eliminating many jobs. The moment leadership presents ROI as headcount avoided, you have confirmed that fear with your own slide deck. Every subsequent conversation about adoption becomes a negotiation about job security.

Measure hours returned per person instead. A quality analyst who reclaims six to ten hours a week on documentation is a defensible business case, and it points somewhere useful: supplier development, capability studies, the CAPA backlog nobody has touched in a year. That framing survives scrutiny from finance and from the floor at the same time. It also does not require a single person on your team to believe AI is good for society, only that their Thursday afternoons get better.

Why Scepticism Is Becoming a Planning Input, Not Background Noise

The noise level around AI is rising from every direction at once. Tech leaders have publicly warned about the threat of AI and called for steps to slow it down before it outsmarts humans. Lawmakers are weighing what, if any, action to take. President Trump has called those warnings a “hoax” while announcing a “Super Intelligence Force” task force, and in September 2026 met tech executives at the White House who signed a “constitution” to police themselves.

None of that resolves anything for your staff. It keeps the subject in the news cycle every week, which means your operators will hear something contradictory about AI between your pilot kickoff and your go-live. The infrastructure fight is moving into local politics too: when roughly two-thirds of a country object to data centres, planning hearings and council votes follow. European manufacturers face the same sentiment pressure with tighter regulation layered on top.

What to put in your 2027 AI plan as a result

Treat workforce trust as a budget line, not a soft factor. Give it hours, an owner, and a milestone the same way you treat validation or integration. If nobody is accountable for it, it gets dropped the first week the schedule slips.

Document your human-in-the-loop controls now, while the system is small and the documentation is cheap. Write down which decisions the model proposes and which a qualified person signs, where the audit trail lives, and what happens when the model is wrong. Retrofitting that evidence onto a live system after a customer audit costs several times more.

• Trust budget: named owner, scheduled sessions with line staff, and a feedback route that produces visible changes.
• Control register: every AI-assisted decision mapped to a human approver and a retention policy.
• Data residency statement: one page, plain language, ready for customers and for your own team.
• Headcount position in writing: what the automation does and does not change about roles.

Expect to justify all four to auditors, customers, and your own people within two years.

Source: abcnews.com