Anthropic
Asking 120,000 people
what they really think of AI.
Anthropic has launched a new initiative called "Inviting hard questions." Through a 90-second film and a dedicated site, it's soliciting the toughest questions about jobs, safety, families, and scientific hope -- and pledging to publish its reasoning in stages. The research behind it spans over 120,000 people. Is this just a brand campaign, or a new template for accountability?
The Initiative
Soliciting the questions
that are hardest to answer
Anthropic unveiled a new brand and research initiative on July 9.
According to Anthropic's announcement, "Inviting hard questions" launched on July 9. Built around a 90-second film and a dedicated submission site, it asks the public for its toughest questions about how AI affects jobs, society, and families, as well as its promise for science and medicine. In return, Anthropic pledges to publish its reasoning and specific actions in stages -- and to be explicit when it falls short of a stated goal.
This didn't come out of nowhere. Over the past year, Anthropic says it has asked more than 120,000 people worldwide about their hopes and fears for AI. That includes a survey of 52,000 Americans, a separate poll of 81,000 Claude users across 159 countries and 70 languages, in-person focus groups, and analysis of real-world usage data. The new initiative builds directly on that base, creating a formal channel for confronting the questions people find hardest to ask.
| Prior efforts | Inviting hard questions |
|---|---|
| Technical collabs like Project Glasswing | Public engagement on social and ethical dilemmas |
| Internal surveys and usage analysis | Public question submissions, built on that research |
| Disclosure via individual announcements | Ongoing publication of reasoning and unmet goals |
By the Numbers
The research behind the campaign
In Practice
What actually happens
It's not just a call for questions -- it's designed as a response-and-verify cycle.
Collect the questions
The dedicated site takes the public's hardest questions spanning jobs, safety, families, and science.
Cross-reference with research
Submissions get checked against the 120,000-person response set and usage data already collected, to surface the highest-priority themes.
Publish actions and reasoning
Specific actions taken, and the reasoning behind them, get published on an ongoing basis.
Disclose the misses too
Anthropic pledges to be explicit, not silent, whenever it falls short of a stated goal.
Before promising answers, it promised to take the questions seriously in the first place.
Why It Matters
Why this matters for the industry
A step beyond technical collaborations, into accountability as a competitive axis.
Anthropic's public communications have mostly centered on technical collaborations like Project Glasswing, or model-performance announcements. This initiative is a step further: an attempt to make visible how a company itself responds to the anxieties AI raises in society. Soliciting questions and publishing the reasoning behind the response reads as a deliberate move to build trust with regulators and everyday users alike. It's a sign that competition among frontier labs is expanding beyond raw performance into how accountability itself gets presented.
If you're in leadership
When explaining your AI adoption policy internally, you can cite Anthropic's published concerns-and-actions table as reference material -- a ready-made template for accountability messaging.
If you're a PM
The published response data works as primary-source material for anticipating what your own users worry about when AI touches your product.
If you're an everyday user
There's now a formal channel to submit the worries you actually have about jobs or family. Whether the responses are followed through on is something you can judge as the disclosures roll out.
What's Next
What happens next, and what to watch
The near-term thing to watch is the substance of what gets published as "reasoning." Whether this stays at the level of abstract principle, or actually drives concrete product changes and policy shifts, will determine how the initiative is judged. Recommended actions: (1) map your own AI-usage policy against the concern categories Anthropic has published -- jobs, safety, families, science; (2) use the published response data in competitive analysis or internal accountability materials; (3) track the timing of the first "actions taken" disclosure.
There's a risk worth naming too. So far, the public call and the film have led -- it's not yet clear when, or at what level of detail, the "unmet goals" disclosures will actually appear. It's reasonable to read this partly as a brand campaign, and the transparency on offer could end up more limited than it sounds. It's already being covered alongside remarks from Sam Altman, the kind of framing that invites comparison with competitors -- so how substantive it really is will depend on what gets published next.