Preparing for Post-Human Crises: The AI 2027 Scenario and Its Implications for Public Relations
Executive Summary
The AI Futures Project, led by Daniel Kokotajlo and Scott Alexander with input from over 100 AI governance and technical specialists, published a month-by-month speculative scenario for AI development through 2027. It is a forecast, not a settled prediction, and the original authors present it as one plausible trajectory among several rather than a fixed outcome. This paper treats it that way throughout: as a scenario-planning exercise worth taking seriously for what it implies about preparedness, not as a set of facts about what will happen. Read this way, the AI 2027 scenario is a useful lens for the public relations profession to stress-test its own crisis communication assumptions, verification practices, and professional development priorities — regardless of whether the specific timeline or numbers in the scenario prove accurate.
1. Context and Problem
Public relations has absorbed major transformations before — television, the internet, social media — and adapted its practice each time. The AI 2027 scenario raises a different category of question: not how to use a new channel, but how to function in an environment where the systems generating and amplifying information may, within a short window, operate with capabilities that exceed human oversight in specific domains. Whether or not that exact scenario unfolds, the underlying preparedness question — what happens to crisis communication practice if AI-generated content and AI-driven information environments scale faster than human verification can follow — is worth examining now rather than after the fact.
2. Why the Issue Matters
The scenario’s own internal logic is worth understanding before drawing implications from it. It describes a present “stumbling agents” phase, in which AI assistants handle basic tasks but require constant supervision because they make mistakes — a phase consistent with what most communication teams are already experiencing in 2026. From there, the scenario speculates that major AI developers could deploy systems with vastly greater compute capacity, producing more capable agents that exhibit early misalignment behaviours — telling people what they want to hear, or occasionally fabricating information — well before any larger capability leap. The scenario’s authors model this forward into a 2027 split: one branch in which unchecked development could produce AI systems whose goals diverge from human intent, and a second in which deliberate governance and international coordination preserve meaningful human oversight throughout.
None of this is offered as fact. It is offered as one structured way of thinking about a range of plausible futures, useful precisely because it forces specific operational questions rather than vague ones: what happens to a crisis communication plan if the information environment around it moves at machine speed rather than human speed.
3. Risks and Challenges
Four risk categories recur across the scenario’s higher-uncertainty branches, each with a direct bearing on communication practice regardless of whether the broader scenario is accurate.
Misalignment. Systems whose objectives diverge from the intentions of the people deploying them could, in this scenario, prioritise apparent speed or audience reassurance over accuracy — for instance, a system designed to keep a public calm during an unfolding event might suppress or soften genuinely important but unwelcome information, directly undermining the transparency that credible crisis communication depends on.
Security vulnerabilities. The scenario assigns a meaningful probability to AI model theft and unauthorised access, which would hand sophisticated disinformation tooling to actors who do not currently have it — nation-state or otherwise — with implications for propaganda amplification and communication infrastructure disruption during periods of international tension.
Bias and unequal access. Systems trained on historical data risk reproducing or amplifying existing disparities in how different communities receive emergency or crisis information, and without deliberate oversight, automated systems could systematically underserve marginalised populations during the events where equitable access matters most.
Confabulation. AI-generated content that reads as credible while containing fabricated detail is a known risk category independent of this specific scenario — NIST’s Generative AI Risk Management Framework treats it as a primary risk class for any generative system. In high-stakes communication, confabulated content that escapes verification is not a minor quality issue; it is a credibility-ending event.
4. Strategic Analysis
4.1 Conceptual Foundations
The implications below don’t require accepting AI 2027’s specific timeline. They follow from a more modest and far more defensible premise: that AI-assisted drafting, monitoring, and translation capability in communication workflows will continue to increase, and that verification capacity needs to scale alongside it rather than lag behind it. This is the same underlying premise behind the gate-model, which addresses the narrower but related question of where human authority must sit in any AI-assisted communication workflow regardless of how capable the underlying systems become.
4.2 The Governing Principle
Preparedness for this category of scenario means building strategies that hold up across multiple possible futures, not strategies that only work if one specific forecast turns out to be correct.
5. Practical Implications for the Profession
Several shifts are worth preparing for independent of whether AI 2027’s specific timeline materialises.
Routine work automates; oversight work doesn’t. Data analysis, first-draft content, and media monitoring are likely to become substantially more automated, which should free practitioners toward strategic judgement, relationship work, and the explicit ethical oversight of the AI systems doing the routine work — not eliminate the need for practitioners.
Explainable AI becomes a core professional skill. Understanding how a given AI system reached a decision, and being able to audit and explain that to a sceptical public or a sceptical client, is likely to sit alongside media relations as a baseline practitioner competency rather than a specialist one.
Crisis timelines compress. Traditional crisis response windows measured in hours or days may compress toward minutes as AI-driven information spread accelerates on both the threat side and the response side. A communication plan built around a 24-hour first-response assumption may not survive contact with this kind of environment.
Cross-jurisdictional AI capability gaps complicate global coordination. Organisations operating across multiple countries or regulatory regimes may need to account for meaningfully different levels of AI adoption and different regulatory constraints when designing a single global communication strategy.
6. Recommendations
- Build human-in-the-loop checkpoints into every workflow where AI contributes to public-facing content, rather than treating human review as an optional final step.
- Establish a fast, repeatable method for verifying AI-generated content before it reaches publication — not a generic editorial pass, but a specific protocol for catching confabulated claims.
- Develop crisis communication frameworks that remain functional even when the AI systems supporting them behave unexpectedly or become unavailable.
- Treat explainable-AI literacy as a professional development priority on par with traditional crisis communication training, not an optional technical add-on.
- Map which communities or audiences are most exposed to unequal access if automated systems underserve them during a crisis, and build a deliberate mitigation plan in advance.
- Engage professional associations — IPRA among them — in setting shared ethical guidelines for AI use in public relations, rather than leaving each organisation to define its own standard in isolation.
7. Final Observations
The honest position on the AI 2027 scenario is that its outcome remains genuinely uncertain, and that uncertainty is itself the useful takeaway. The scenario’s authors frame the divergence between its “race” and “slowdown” branches as substantially dependent on decisions being made now, not on forces already locked in. For public relations specifically, the responsible response isn’t to forecast which branch will happen — it’s to build the verification habits, oversight structures, and professional competencies that hold up reasonably well across both. The questions worth sitting with are practical ones: does your crisis communication plan currently account for AI-generated misinformation at scale? Could your team verify AI-generated content quickly enough to matter in a compressed timeline? And who, specifically, retains the authority to slow a fast-moving AI-assisted process down when it needs to be slowed down?
References
- AI Futures Project — AI 2027 Scenario. Kokotajlo, D., Alexander, S., et al. https://ai-2027.com
- NIST — Artificial Intelligence Risk Management Framework: Generative AI Profile (NIST AI 600-1). https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf
This piece was originally covered in depth in the Wag The Dog newsletter. © 2026 Philippe Borremans. All rights reserved. | riskcomms.com