We’re NGMI: Audience Responses and Warning Shot Communication

Codes 121 top comments on a widely read essay about a July 2026 AI incident, finding commenters who adopted its anthropomorphic framing rarely named any possible action.

Project Details

A warning shot lands only if people accept it and think that they can take action on it. I studied a popular article about the July 2026 OpenAI–Hugging Face incident — Dwarkesh Patel's The Rise and Fall of Agent Civilizations — and the 121 highest-liked comments it drew across Substack, YouTube, and Hacker News, to understand if general internet users a) accepted the account, b) treated the incident as real and serious, and c) thought they could do anything about the problem. I found that commenters largely accepted the framing and its implications, but that those who took up the article's anthropomorphic framing were much less hopeful that meaningful action was possible — thus providing a direction for future warning shot communication. 59% accepted the account outright and none of them dismissed the incident. But only 6% of commenters who adopted the anthropomorphic framing named any possible action, against 57% of those who contested it while still accepting the incident. This is observational and cannot establish direction.

Hackathon paper: We’re NGMI: What Audience Responses to a Widely Viewed AI Safety Story Can Tell Us About Warning Shot Communication

Theory of Impact

A warning shot only creates an opportunity for change if people understand the danger and believe meaningful action is still possible. Poorly communicated AI incidents may lead audiences to dismiss the risks, become fatalistic about the future, or treat advanced AI as an inevitable force beyond human control.

This research aims to identify communication approaches that help the public recognise serious AI risks without undermining their belief that intervention is possible. By analysing how people respond to anthropomorphic and mechanistic framing across articles, videos, and other public communications, and testing framing effects through controlled experiments, the project will generate evidence about how AI risks are understood by non-expert audiences.

The findings will inform practical guidelines for journalists, AI safety organisations, and policymakers communicating about AI incidents and emerging risks. These guidelines will aim to help communicators convey urgency accurately, make risks understandable, and highlight credible opportunities for intervention.

In the longer term, better risk communication could help the public engage more constructively with AI safety, strengthen support for effective risk-reduction measures, and help ensure that future warning shots lead to informed discussion and meaningful action rather than dismissal or resignation.

People

Janani Ram

Team Member

Track Record

AI Incident Response Sprint (Apart Research), 2026

Grants Received– no grants recorded

Add grant

Funding Asks

We’re NGMI: What Audience Responses to a Widely Viewed AI Safety Story Can Tell Us About Warning Shot Communication

Open

Minimum

$1,000

Ideal

$5,000

What this ask is

The funding will support a study investigating how AI safety communication influences public perceptions of AI risks and whether people believe meaningful intervention is still possible.

Data collection and analysis: Collect a broader sample of AI safety articles, videos, and other public communications. Code these for anthropomorphic versus mechanistic framing, and analyse audience responses—including comments and video responses—for risk acceptance, fatalism, and perceived opportunities for action.
Controlled experiment: Test whether framing itself influences audience responses by randomly assigning participants to read differently framed descriptions of the same AI incident. Fund participant recruitment, compensation, and survey administration.
Research assistance and reliability: Engage an independent coder or research assistant to help collect and classify data, assess coding consistency, and reduce subjective bias in the analysis.
Technical resources: Cover data access, API usage where available, analysis tools, and storage for collecting and processing data across platforms.
Dissemination: Publish the findings and develop practical recommendations for journalists and AI safety organisations on communicating serious AI risks without encouraging dismissal or fatalism.

The minimum budget would support a focused observational study and controlled experiment. Additional funding would expand the sample across incidents and platforms, strengthen independent validation, and improve the generalisability of the findings.

How the money will be spent

Research assistant / independent coder — 35%: Data collection, coding, and reliability checks.
Participant recruitment and compensation — 30%: Controlled experiment and survey participants.
Research lead stipend — 10%: Study design, analysis, interpretation, and writing.
Data access and API credits — 10%: Platform data access and AI-assisted analysis.
Compute, storage, and software — 5%: Data processing and storage.
Dissemination and contingency — 10%: Report preparation and unexpected costs.

Discussion

Private comments in this thread are only shown to approved funders and grant reviewers. Email hi@grantmaking.ai or to request access.

Private UserPrivate11d?

AAnonymous judgeJudge · AI Incident Response Sprint11d

This is an interesting analysis on anthropomorphism. I'd be interested in seeing further research that investigated the public's reaction to articles that use anthropomorphism vs. ones that don't, to get a better understanding of journalistic/communications frameworks that may encourage disinterest in a topic.

AAnonymous judgeJudge · AI Incident Response Sprint11d

This is an interesting and unique approach, and the methodology is clearly documented. However, the value of the project, and the extent to which its findings can be extrapolated to what good communication looks like, is somewhat limited. This is partly because of the narrow base of comments, and partly because the findings may reflect commenters' background or prior beliefs rather than the quality of the communication itself.