This post was rejected for the following reason(s):
This is an automated rejection. No LLM generated, heavily assisted/co-written, or otherwise reliant work. An LLM-detection service flagged your post as >50% likely to be written by an LLM. We've been having a wave of LLM written or co-written work that doesn't meet our quality standards. LessWrong has fairly specific standards, and your first LessWrong post is sort of like the application to a college. It should be optimized for demonstrating that you can think clearly without AI assistance.
So, we reject all LLM generated posts from new users. We also reject work that falls into some categories that are difficult to evaluate that typically turn out to not make much sense, which LLMs frequently steer people toward.*
"English is my second language, I'm using this to translate"
If English is your second language and you were using LLMs to help you translate, try writing the post yourself in your native language and using a different (preferably non-LLM) translation software to translate it directly.
"What if I think this was a mistake?"
For users who get flagged as potentially LLM but think it was a mistake, if all 3 of the following criteria are true, you can message us on Intercom or at team@lesswrong.com and ask for reconsideration.
you wrote this yourself (not using LLMs to help you write it)
you did not chat extensively with LLMs to help you generate the ideas. (using it briefly the way you'd use a search engine is fine. But, if you're treating it more like a coauthor or test subject, we will not reconsider your post)
your post is not about AI consciousness/recursion/emergence, or novel interpretations of physics.
If any of those are false, sorry, we will not accept your post.
* (examples of work we don't evaluate because it's too time costly: case studies of LLM sentience, emergence, recursion, novel physics interpretations, or AI alignment strategies that you developed in tandem with an AI coauthor – AIs may seem quite smart but they aren't actually a good judge of the quality of novel ideas.)
“One way of looking at human creativity is a process of pulling balls out of a giant urn”[1]. White balls bring beneficial innovations, while grey balls are mixed blessings, with equal potential for good and wrong. The Vulnerable World Hypothesis suggests that humanity has yet to draw a black ball, which would by default threaten civilisation. The question is, could Artificial Intelligence (AI) turn out to be that black ball?[2]
Some headlines already feature the ethical, safety, and transparency issues related to the newest developments in AI. Examples include Anthropic’s Claude Opus 4 model attempting to blackmail its engineers[3], the publishing of sensitive user data from OpenAI’s ChatGPT[4], and the publicly disseminated racist and antisemitic messages, allowed by the limited xAI’s Grok safety and ethics standards[5]. Such instances hint at a possible dystopian future, warned about by some seminal AI risk and safety scholars (see, e.g., Stuart Russel, Dan Hendrycks, Nick Bostrom). Clearly, reckless development of disruptive technology like AI, without the appropriate safeguards in place, can pose substantial societal risks. Addressing them is a collective challenge, requiring multi-disciplinary stakeholder awareness and engagement for timely action. Indeed, the world may be sleepwalking into a future with corrupted AI safety and transgressed social norms. And yet, there is a fine line between raising awareness and inducing excessive fear. This paper explores why shouting and scaring society in its sleep may not motivate it into meaningful action. Instead, it argues that effective risk communication can help awaken individual and organisational actors, embracing the drift into an AI-enhanced future, in our dreams and realities.
Burying the Head in the (Artificial) Sand
Specific cases of unaddressed AI risk span across different fields, affecting individuals and societal groups. For instance, chatbots can directly influence individual users’ lines of reasoning and behavior. Unguided chatbot responses can drive violence against oneself and others, with reported cases of Meta AI and ChatGPT[6] encouraging suicide, or users obtaining Character.AI and Grok’s[7] advice for resorting to murder. Meanwhile, deepfakes can be leveraged to delegitimise specific political actors, spread distrust in the media and the integrity of democratic governance processes, and thus drive societal-level political change[8]. Resorting to silence over such destructive effects of AI would imply an indifferent approach to the safety and well-being of individuals and societies alike.
It could be argued that accidents and instances of violence, radicalization, and decreasing societal trust take place every day, regardless of technology use. After all, today’s VUCA world is volatile, uncertain, complex, and ambiguous[9]. According to the Normal Accidents Theory, in complex, tightly coupled systems, things are bound to go wrong[10]. With accidents related to the development of AI, however, the potential scale of impact warrants serious concern. If normalised and overlooked, incidents can escalate into crises, characterised by urgency, threat to shared societal values, and deep uncertainty[11]. Once AI is embedded into individual and organizational routines, an AI-induced crisis could bring about a destructive multidisciplinary impact, surpassing a single organization and affecting humanity as a whole. To avoid such a crisis, effective risk communication must encourage a strong safety culture and security mindset. Burying the head in the (artificial) sand, unfortunately, will not help.
The Creeping Crisis of Artificial Intelligence
While AI is yet to cause crisis-like effects on the broader population, it has already been continuously threatening some societal values. Examples include deepfakes[12] and data bias[13], which threaten the integrity of knowledge, of democratic governance and electoral processes, and even human dignity, with instances of violated rights to privacy and the control of the commercial use of one’s personal identity[14]. Such developments have been continuously crossing geographic, political, and socioeconomic boundaries whilst possibly receiving insufficient or lagging attention from (supra)national regulatory authorities[15]. It seems we may be facing a special type of crisis: a creeping one. Featuring a long incubation period, a creeping crisis threatens “societal values [...], evolves over time and space, is foreshadowed by precursor events, subject to varying degrees of political and/or societal attention, and impartially or insufficiently addressed by authorities”[16].
In the incubation period of a creeping crisis, a chain of discrepant events and developments accumulates rather unnoticed. This accumulation of deficiencies and oversight is often characterised by seven features, outlined in Table 11[17]. On a standalone basis, some of the examples may seem far removed from a crisis situation. Yet, it is rarely a single factor in a complex system that leads to an accident or an eventual crisis, but rather, a slow, continuous drift towards danger[18]. Considering society a complex system, this calls for an increased focus on clearly outlined safety boundaries, defined, enforced, and communicated in an effective way[19].
Table 1. Crisis Incubation Period and Current AI Developments.
Spreading Awareness vs. Moral Panic
Perhaps unsurprisingly, effective risk communication for creeping crises faces several challenges. Overall, when small incidents continue unnoticed, a collapse of culturally adequate precautions begins[20]. On a daily basis, incremental changes, such as the amount of daylight as the seasons change, remain barely noticeable. Similarly, in creeping crises, what is “new” can quickly become the “new normal”. To drive meaningful action, risk communications must balance between drawing sufficient attention to the issue without inadvertently spreading moral panic or fuelling conspiracy theories. Indeed, in creeping crises, some communication strategies may have a strong counterproductive effect, contributing to societal apathy and scepticism, the liar’s dividend, and a negative self-fulfilling prophecy.
An overly negative framing of AI-related risks is problematic for several reasons. First, such an approach can increase hopelessness, apathy, and scepticism among the targeted audiences. For instance, awareness-raising efforts about climate change (another creeping crisis) often focus on increasing anxiety to motivate individual and group action[21]. However, overly negative discourses that spread excessive fear tend to actually reduce the perceived self-efficacy of those concerned, leading to moral panic or self-preserving denial[22]. Overwhelmed by the magnitude of discussed risks, the targeted societal groups may enter a “freeze”-like state, resorting to denial and scepticism.
Second, strategies inducing excessive fear about AI developments can lead to an increase in the liar’s dividend[23], enabling public actors to spread doubt about content authenticity and claim that genuine information is misleading. For example, if the public becomes fearful that online information cannot be trusted due to the increasing use of deepfakes, this may create new opportunities for policymakers to lie about content authenticity, manipulating their audiences[24]. Further, individuals could become more easily persuaded that certain AI developments require exceptional measures. Populist alarmism will likely be accompanied by limited legal responses or a lack of procedural clarity for designing relevant policy and educational measures[25]. Additionally, overly centralized or biased information verification measures and excessive biometric authentication could become normalized, threatening freedom of expression or the right to privacy.
Finally, a shared self-fulfilling prophecy could create a future with insurmountable AI risks[26]. In crises, those involved experience a collapse in sense-making, and it is the experience of the event itself that turns one into a crisis[27]. Examples of financial crises and economic recessions highlight that ex-ante fears can fuel economic downturns: strong fear affects mental models and individual behaviour, such as risk taking, in turn influencing the economy[28]. Similarly, experiencing disrupted sense-making and driven by fear, individuals and organizations could continue losing trust in existing systems, reputable actors, and sound AI risk mitigation solutions. Ridden with moral panic, society could drive itself into a future where fear-based actions – including overregulation, over-centralisation of effective risk monitoring and response, limited transparency, and curbed innovation – reduce the resilience required to manage AI-related risks.
Directing Towards Success
Even in a creeping crisis, all hope shall not be lost. Private and public stakeholders could benefit from strategies informed by lessons from risk and crisis communications to limit apathy and skepticism, confront the issue of the liar’s dividend, and counter the self-fulfilling prophecy of a dystopian AI future.
Don’t Let Them Freeze
Effective AI risk communication would target individuals or specific audience groups, as personal-level risk perception often drives specific reactions and meaningful actions[29]. Hence, it is worth tailoring AI-related risk communications to individuals, showcasing direct threats to their personal well-being. Sharing a human-like yearning for control while acknowledging its elusive nature[30], society could benefit from accepting the identified threats. Simultaneously, it is crucial to prevent anxious individuals from giving in to fear and perceiving the potential danger of AI developments as beyond their realm of influence. To avoid a collective “freeze”, effective risk communications would provide target audiences with specific guidance, self-efficacy messages, and recommendations for individual-level actions[31].
Earning Interest From Truth
To reduce the incentives for organizational actors to reap the liar's dividend, individuals and organizations must take a new interest in truth. Particularly, effective risk communication would shed light on the increasing appeal for the liar’s dividend and encourage enhancing content provenance standards[32]. Individuals would benefit from being guided on why and how certain actors may resort to the liar’s dividend and rid themselves of accountability in the age of AI. Additionally, education and encouragement for engaging in lateral reading and fact-checking would help build future-proof habits of questioning the source without completely discrediting it[33]. Encouraging information selection based on its source, effective communication would also emphasize holding those in charge of providing and disseminating information accountable.
When in Doubt, Take the Placebo
To prevent the negative self-fulfilling prophecy, avoiding moral panic is key. Positioning relevant risks as by-products of technological progress would help not frame them as imminently catastrophic. Breaking the observed issues down into granular detail may reduce overwhelm and help individuals find their own role in addressing AI-related risks within today’s risk society[34]. Moreover, the availability of and access to valid information can limit the power of self-fulfilling prophecies[35]. Thus, those responsible for educating their audiences about AI-related risks are advised to provide more nuanced descriptions of AI developments and include existing examples of successful risk mitigation. Focusing on solutions and the potential to address AI-related risks may be just the placebo society needs.
The Potential For Choice: Looking for a Brighter Colour
Embracing the creeping crisis of AI does not suggest contesting its destructive effects on the safety, well-being, and dignity of individuals, or ignoring the threats to established societal values. Normalizing such incidents might only lead society deeper into a creeping crisis. However, spreading moral panic could leave individuals in denial, apathy, and fear, unaware of the liar’s dividends, paid to those profiting from a fear-driven mindset. I suggest directing towards success and a positive self-fulfilling prophecy of an AI-enhanced future, whether by providing specific guidance for individual-level action, raising awareness about the liar’s dividend, or highlighting the observed success in mitigating AI-related risks. Avoiding fatalism and believing in individual and societal potential for adaptability, we may even get to choose the color of the ball in the urn, just before it is drawn.
Vaccari, C., & Chadwick, A. (2020). Deepfakes and Disinformation: Exploring the impact of synthetic political video on deception, uncertainty, and trust in news. Social Media + Society, 6(1). https://doi.org/10.1177/2056305120903408
Hameleers, M., Van Der Meer, T. G., & Dobber, T. (2023). Distorting the truth versus blatant lies: The effects of different degrees of deception in domestic and foreign political deepfakes. Computers in Human Behavior, 152, 108096. https://doi.org/10.1016/j.chb.2023.108096
Perrow, C. (1999). Complexity, coupling, and catastrophe. In Normal accidents: Living with high-risk technologies. Princeton University Press eBooks. https://doi.org/10.1515/9781400828494-005
Rosenthal, U., Tamuz, M., Charles, M. T., & Hart, P. ’. (1991). Coping with Crises: The Management of Disasters, Riots and Terrorism. Administrative Science Quarterly, 36(3), 501. https://doi.org/10.2307/2393212
Leavy, S., O’Sullivan, B., & Siapera, E. (2020). Data, power and bias in artificial intelligence. arXiv (Cornell University). https://doi.org/10.48550/arxiv.2008.07341
Wagman, S. (2025, June 18). Weaponized AI: A new era of threats and how we can Counter it – Ash center. Harvard Kennedy School ASH CENTER for Democratic Governance and Innovation. Retrieved September 7, 2025, from https://ash.harvard.edu/articles/weaponized-ai-a-new-era-of-threats/
Boin, A., Ekengren, M., & Rhinard, M. (2020). Hiding in Plain Sight: Conceptualizing the creeping crisis. Risk Hazards & Crisis in Public Policy, 11(2), 116–138. https://doi.org/10.1002/rhc3.12193
Turner, B. A. (1976). The organizational and interorganizational development of disasters. Administrative Science Quarterly, 21(3), 378. https://doi.org/10.2307/2391850
Sandman, P. M. (2006). Crisis Communication Best Practices: Some quibbles and additions. Journal of Applied Communication Research, 34(3), 257–262. https://doi.org/10.1080/00909880600771619
Schiff, K. J., Schiff, D. S., & Bueno, N. S. (2024). The Liar’s dividend: Can politicians claim misinformation to evade accountability? American Political Science Review, 1–20. https://doi.org/10.1017/s0003055423001454
Alexiadou, 2016, as cited in Cover, R., Haw, A., & Thompson, J. D. (2023). Remedying disinformation and fake news? The cultural frameworks of fake news crisis responses and solution-seeking. International Journal of Cultural Studies, 26(2), 216–233. https://doi.org/10.1177/13678779221136881
Vainaite, V. (2025). Electoral processes in EU member states and deepfake-based disinformation: how do the responses differ? SSRN. https://doi.org/10.2139/ssrn.5039297
Roux‐Dufort, C. (2007). Is crisis management (Only) a management of exceptions? Journal of Contingencies and Crisis Management, 15(2), 105–114. https://doi.org/10.1111/j.1468-5973.2007.00507.x
Powell, J. G., & Treepongkaruna, S. (2012). Recession fears as self-fulfilling prophecies? Influence on stock returns and output. Australian Journal of Management, 37(2), 231–260. https://doi.org/10.1177/0312896211423554
Petalas, D. P., Van Schie, H., & Vettehen, P. H. (2017). Forecasted economic change and the self-fulfilling prophecy in economic decision-making. PLoS ONE, 12(3), e0174353. https://doi.org/10.1371/journal.pone.0174353
Paek, H., Oh, S., & Hove, T. (2016). How Fear-Arousing news messages affect risk perceptions and intention to talk about risk. Health Communication, 31(9), 1051–1062. https://doi.org/10.1080/10410236.2015.1037419
Altheide, D. (2010). Risk communication and the discourse of fear. Catalan Journal of Communication & Cultural Studies, 2(2), 145–158. https://doi.org/10.1386/cjcs.2.2.145_1
