Narratives and false claims divide the public debate on AI risks

The public AI risk debate has differentiated sharply since spring 2023. In March 2023, the Future of Life Institute published an open letter calling for a minimum six month moratorium on training AI systems more capable than GPT-4. Prominent signatories included Elon Musk, Steve Wozniak, Yoshua Bengio, Stuart Russell and Yuval Noah Harari ⁵. On 30 May 2023, the Center for AI Safety issued a statement saying that reducing existential risk from AI should be a global priority alongside pandemics and nuclear war.

Warnings from research and industry

It was signed by Geoffrey Hinton, Yoshua Bengio, Demis Hassabis (DeepMind) and corporate leaders Sam Altman (OpenAI) and Dario Amodei (Anthropic), among others ⁸+224 TIME25 Science Media Centre. TIME Magazine classified the statement on the same day as a rare collaboration among competing corporations ²⁴, while the British Science Media Centre simultaneously collected expert reactions ranging from clear approval to sharp criticism of the escalation ²⁵.

In parallel, a counter-movement developed that criticised the shift of public attention to hypothetical future scenarios and instead foregrounded current harms. Already in June 2023, an article by The Conversation France argued that concrete risks should be addressed before potential existential threats ¹⁴. Scientific American noted in July 2023 that AI was indeed an existential threat, but not in the form of an awakening superintelligence, rather through the gradual erosion of democratic processes and opinion formation ²⁶.

Critics highlight concrete present-day harms

An essay in Neue Gesellschaft/Frankfurter Hefte from September 2023 summarised the criticism. Science fiction narratives distracted from the real challenges and opportunities of the technology ³¹. In December 2023, a commentary from RAND Corporation described philosophical debates about AI extinction risks as distracting from already existing problems ¹. Princeton authors Arvind Narayanan and Sayash Kapoor called numerous AI promises AI Snake Oil in December 2024 and drew a sharp distinction between marketing messages and reliable capabilities ³³.

In several contributions, longtermism is described as the intellectual framework of the extinction debate. It is a philosophical movement formulated in the milieu of the Future of Humanity Institute in Oxford around Nick Bostrom and William MacAskill. Critical observers such as Timnit Gebru and Émile P. Torres summarise the related schools of thought under the acronym TESCREAL (Transhumanism, Extropianism, Singularitarianism, Cosmism, Rationalism, Effective Altruism, Longtermism). Asterisk magazine critically debated the analytical value of this umbrella term in April 2024, while Reason Magazine accused the EA movement of an authoritarian side in its approach to AI regulation in July 2024 ²⁰+132 Asterisk Magazine.

An article published in AI & Society (Springer Nature) in February 2026 classifies alarm rhetoric as a pattern of moral panic and proposes a shift towards pragmatic management of AI effects in work, education and ethics ⁴⁶.

On the side of the warning voices, institutional senders such as the Future of Life Institute and the Center for AI Safety stand out, along with individual figures from research and business, including Geoffrey Hinton, Yoshua Bengio, Stuart Russell, Yuval Noah Harari, Elon Musk and Steve Wozniak. At the same time, corporate executives such as Sam Altman (OpenAI), Dario Amodei (Anthropic) and Demis Hassabis (DeepMind) signed the CAIS statement ⁵+28 All About Security24 TIME. In November 2025, a senior figure from the United Nations, the UN special envoy for new technologies, spoke out and classified advanced AI as a possible existential risk for humanity to franceinfo ³⁴.

As funders of the AI safety and longtermism scene, Open Philanthropy (Dustin Moskovitz and Cari Tuna), Jaan Tallinn and FTX founder Sam Bankman-Fried, who was convicted of fraud, are cited ⁴. The amount of funds that are said to have been spent within the Effective Altruism sphere on AI existential risks is disputed. Netzpolitik.org speaks in its analysis of approximately 1.5 billion US dollars ⁴, Critic Magazine picks up the figure in a critical article ¹¹, while other available sources do not confirm the figure. An article at Yahoo News describes for the year 2026 a targeted intimidation campaign by AI billionaires against critics, regulators and critical reporting. The author classifies the campaign as already effective ³⁹.

Critical positions of leading researchers

On the side of critics stand prominent expert voices such as Meta AI chief Yann LeCun, who called concerns about AI existential risk "complete B.S." to TechCrunch on 12 October 2024 and rejected the thesis that artificial intelligence is about to become truly intelligent ²⁹. Princeton researchers Arvind Narayanan and Sayash Kapoor argue similarly with strong expert backing against exaggerations ³³. The libertarian and regulation-sceptical Reason Magazine rejected on 11 September 2026 the thesis that AI doom is a targeted psyop, but considers the discourse to be substantively misdirected ³⁷.

The equally market-liberal Abundance Institute describes in a paper from August 2025 AI techno-panic as a separate harm to innovation and policy ⁴⁸. Arguing from an expert perspective, TechTarget demands in January 2026 a way beyond doomerism that cleanly separates hype and reality ³⁸. The media-critical portal Übermedien describes in November 2025 how German reporting amplifies fear of fearmongering itself ⁴⁴, and Business Punk sees a way out in October 2025 primarily in building competence rather than further outrage ⁴⁹.

Fact-checkers and media research provide empirical counter-evidence. Correctiv published on 23 December 2025 an overview of the defining narratives and false claims of 2025 ²¹ and on 3 August 2026 a practical test of how easily news can be faked with ChatGPT and other systems ²². A Correctiv background piece from 28 April 2026 also documents how deceptively authentic, AI-generated women on TikTok conduct opinion-making for the AfD on a large scale and partly in erotic presentation ⁴³.

Snopes maintains a constantly updated collection of AI-related rumours and fabrications ²³ and published on 30 December 2025 an annual review of approximately 25 debunked AI slop rumours ⁴². The EU DisinfoLab, an independent expert association against disinformation based in Brussels, has been operating a continuously updated AI disinformation hub for European cases of AI-supported disinformation since April 2026 ⁴⁵. Euronews compiled practical guidance on International Fact-Checking Day in April 2026 on how to identify AI-generated disinformation ⁴¹.

The Harvard Kennedy School Misinformation Review documented in June 2026 an expert survey in which experts disagree on countermeasures, but unanimously demand separate approaches for health and politics respectively ⁴⁷. The Knight First Amendment Institute at Columbia University examined 78 election deepfakes and concluded that political disinformation is at its core not an AI problem because the impact of AI-generated content remained limited ³⁰.

Surveys on public perception

On the survey side, the Pew Research Center ¹⁷+118 Pew Research Center with worldwide and US surveys from autumn 2025, the Reuters Institute with its Digital News Report Generative AI 2025 ¹⁹ and German Bitkom ¹⁶ with a survey from February 2026 appear. New additions include a YouGov Germany analysis from May 2026, according to which Germans rate the role of AI over the coming ten years more often than average negatively and lie internationally at the pessimistic end ⁴⁰, as well as a survey published on behalf of Berliner Zeitung from September 2026, according to which one in three Germans is afraid of the consequences of AI, with particularly high values in eastern Germany ³⁶.

On the trust picture of media, medienrot.de provides a survey in February 2026 that attributes the highest trust to public broadcasters in handling AI in journalism ⁵⁰. The University of Zurich presented a large-scale experimental study on the perception of AI risks in April 2025 ²⁷. The Reuters analysis of the media role from May 2024 remains the most important reference point for criticism of industry closeness in reporting ¹⁰.

Specialist media such as netzpolitik.org ⁴+15 netzpolitik.org, heise online ⁷, MIT Technology Review ⁹, 1E9 ¹³, Asterisk ³², Reason ²⁰+137 Reason, Abundance Institute ⁴⁸, TechTarget ³⁸, Business Punk ⁴⁹ and Wissenschaftskommunikation.de ¹² continuously classify the debate. The Handelsblatt articles ³+26 Handelsblatt15 Handelsblatt provide the business press perspective, a market-liberal view of reactions from Silicon Valley and of cultural amplifiers. The Terminator is, according to The Conversation from October 2024, even 40 years after its cinema release, considered a defining image of public AI perception ²⁸. The Qatari state broadcaster Al Jazeera took apart the AI alarm cycle internationally in its media magazine The Listening Post in February 2026 and described it as much talking with little action ³⁵.

In autumn 2025, the debate has once again gained visibility. Handelsblatt described on 10 November 2025 how Silicon Valley has responded to current doomsday warnings ³, and had already noted in an essay on 1 September 2024 that the tech elite are competing with each other in catastrophe scenarios ¹⁵. Two days after the November article, All About Security asked how seriously the new debate about AI existential risks should be taken when the creators of the systems themselves warn ⁸. heise online framed the debate on 14 November 2025 as a discussion between apocalypse and new beginnings ⁷, whilst the magazine 1E9 explored on 5 November 2025 the question of whether AI could actually wipe out humanity by the end of the decade ¹³.

On 18 November 2025, Übermedien warned against reporting that intensifies fear of AI by mixing it with fear of fear-mongering itself ⁴⁴. On 24 November 2025, the UN Special Envoy for Emerging Technologies gave the warning institutional weight by naming a possible existential risk from advanced AI when speaking to franceinfo ³⁴. On 15 December 2025, MIT Technology Review found that AI doomsayers were not giving up and were maintaining their position despite criticism ⁹.

At the turn of the year, Snopes published a summary showing that around 25 AI-related false rumours that had been debunked through fact-checking had found wide circulation in 2025 ⁴².

In winter and spring 2026, new attempts to bring order came to the fore. TechTarget outlined on 15 January 2026 a way beyond doomerism that distinguishes between hype and actual risk ³⁸. AI & Society (Springer Nature) published on 10 February 2026 an article that classifies alarm rhetoric as moral panic and demands pragmatic governance in work, education and ethics ⁴⁶. The medienrot.de portal reported on 17 February 2026 about a survey on media usage and AI in journalism, according to which public broadcasters enjoy the most trust when handling AI ⁵⁰. Al Jazeera's The Listening Post summed up the AI alarm cycle internationally on 22 February 2026 and described it as much talk but little action ³⁵.

Solidification of criticism of the hype

Criticism of the hype has become entrenched. Brookings demanded harder questions in October 2024 about how AI is aligned with human values and warned about simultaneous hype and actual harm ². The article from Wissenschaftskommunikation.de from November 2024 discussed how to talk about AI without fuelling alarmism ¹². The TechCrunch interview with Yann LeCun from 12 October 2024 provided a particularly sharp rejection of existential fear from the heart of the industry ²⁹.

Princeton researchers clarified a few weeks later with AI Snake Oil the difference between marketing message and actual capability ³³. Business Punk argued on 14 October 2025 for building competence instead of panic ⁴⁹. The libertarian Abundance Institute had already described AI technopanic in August 2025 as an independent political and economic harm ⁴⁸. On 11 September 2026, the similarly libertarian Reason Magazine rejected the conspiracy-adjacent thesis that AI doom is a deliberate psyop, but considers the discourse substantively misdirected ³⁷.

An article in Yahoo News from 20 August 2026 took a different emphasis and described an intimidation campaign by AI billionaires against critics and regulators that was already having effect in 2026 ³⁹. A central accusation is that the companies' doomsday warnings serve to distract from regulation by simultaneously overstating the capabilities of their own products and directing political attention toward hypothetical future risks ¹+62 Brookings5 netzpolitik.org31 Neue Gesellschaft Frankfurter Hefte33 Princeton University35 Al Jazeera (The Listening Post)38 TechTarget.

Surveys reveal broad public scepticism

For the first time, reliable surveys are now available. The Pew Research Center published a US study in September 2025 and an international survey in October 2025. In the USA, a majority say they are more concerned than enthusiastic about increasing AI use in everyday life, and 76 per cent doubt that companies will use AI responsibly ¹⁸. The international survey covering 25 countries shows Germany in the middle range with 55 per cent concerned, whilst emerging markets such as India and Indonesia remain mostly optimistic ¹⁷.

The Reuters Institute Digital News Report on generative AI from June 2025 demonstrates widespread scepticism towards AI-produced news and low trust in editorial AI use ¹⁹. The German Bitkom survey from 25 February 2026 sums up the ambivalence. People want AI but at the same time are afraid of it ¹⁶. YouGov Germany published on 27 May 2026 that Germans rate the role of AI in the coming decade more negatively than average and stand at the pessimistic end internationally ⁴⁰.

A survey published on behalf of Berliner Zeitung on 4 September 2026 found that one in three Germans are afraid of the consequences of AI and that concern is particularly high in eastern Germany ³⁶. An experimental study by the University of Zurich with over 10,000 participants from the USA and Great Britain concluded in April 2025 that present AI risks such as discrimination, disinformation and job loss concern respondents significantly more than apocalyptic future scenarios ²⁷.

There are further concrete examples of viral false claims and the role of generative AI in disinformation. Correctiv summarised the defining narratives of 2025 on 23 December 2025 ²¹. A Correctiv test from 3 August 2026 showed that ChatGPT was most easily misused for creating fake news, including deceptively authentic screenshots of Tagesschau articles. Meta AI was most likely to refuse to create them, whilst Microsoft's Copilot labelled outputs with an AI-Generated label ²².

A Correctiv background piece from 28 April 2026 shows how AI-generated women in TikTok channels are advertising the AfD on a large scale and sometimes in erotic format ⁴³. Snopes maintains an open collection of AI-related rumours, including deepfakes and quotes taken out of context ²³, and published a year-end summary on 30 December 2025 of around 25 debunked AI slop rumours ⁴². The EU DisinfoLab has been consolidating European cases of AI-enabled disinformation in a continuously updated AI-Disinfo Hub since April 2026 ⁴⁵.

Euronews collected practical tips for International Fact-Checking Day on 2 April 2026 on how to identify AI-generated disinformation in everyday life ⁴¹. The Harvard Kennedy School Misinformation Review published an expert survey on 4 June 2026 according to which experts disagree about remedies against AI disinformation but agree that health and politics each require their own approaches ⁴⁷. The Knight First Amendment Institute at Columbia University analysed 78 AI-generated election deepfakes and concluded in December 2024 that political disinformation is fundamentally not an AI problem because affected voters were hardly swayed in their opinions by the fakes ³⁰.

Cultural amplifiers are backed by an additional source. The Conversation noted on the 40th anniversary of Terminator in October 2024 that the film continues to shape how an AI threat might look ²⁸. Together with the Handelsblatt article ⁶ and the Frankfurter Hefte analysis ³¹, the observation emerges that images of HAL 9000, Terminator and Ex Machina play a disproportionate role in public perception of AI.

The available sources paint a picture of a polarised but now stable standoff. On one side there is a milieu that emphasises existential long-term risks and is institutionally well networked, with prominent spokespersons from research and industry as well as substantial funding from the sphere of Open Philanthropy, Jaan Tallinn and, until the collapse of FTX, Sam Bankman-Fried ⁴+25 netzpolitik.org8 All About Security. On the other side there is emerging criticism that points to present harms and a possible regulatory distraction through future scenarios ¹+82 Brookings14 The Conversation France26 Scientific American31 Neue Gesellschaft Frankfurter Hefte33 Princeton University35 Al Jazeera (The Listening Post)38 TechTarget46 AI & Society (Springer Nature).

The evidence for strategic use of doomsday rhetoric by corporations is documented multiple times independently, though as a critical finding rather than as proven intent ¹+62 Brookings5 netzpolitik.org31 Neue Gesellschaft Frankfurter Hefte33 Princeton University35 Al Jazeera (The Listening Post)38 TechTarget. A Yahoo News contribution intensifies this finding by describing, for 2026, a targeted intimidation campaign by AI billionaires against critics and regulators ³⁹, whereas the libertarian Reason Magazine rejects the conspiracy-leaning escalation of a deliberate psyop and instead classifies the discourse as substantively misguided ³⁷. The reliability of concrete figures on funding for the field is limited. The sum of roughly 1.5 billion US dollars is mentioned by two contributions but is disputed within the source landscape and should not be passed on as a secure value ⁴+111 The Critic Magazine.

There is now a significantly better data situation regarding public sentiment. Pew, Reuters Institute, Bitkom, YouGov, University of Zurich and the Berliner Zeitung survey consistently document widespread scepticism without populations primarily orienting themselves towards apocalyptic scenarios. Concerns focus more on everyday impacts (employment, relationships, creativity, disinformation) than on extinction fantasies ¹⁶+617 Pew Research Center18 Pew Research Center19 Reuters Institute for the Study of Journalism27 Universitaet Zuerich36 Berliner Zeitung40 YouGov Deutschland. Germany ranks internationally more towards the pessimistic end, the eastern part of the republic is particularly sceptical ³⁶+140 YouGov Deutschland. The survey data thus tends to support the critic position that doomsday discourse takes place primarily in elite public spheres whilst populations have more concrete concerns.

Regarding the role of generative AI in disinformation, the Correctiv test shows that the technical misuse threshold is low ²², the Snopes annual review and Correctiv background on AI women for the AfD on TikTok provide concrete case examples in scope and direction ⁴²+143 Correctiv, and the AI Disinfo Hub of the EU DisinfoLab as well as the Euronews guidance document European cases and identification tips ⁴¹+145 EU DisinfoLab. At the same time empirical evaluations such as the Knight study of 78 election deepfakes relativise the causal effect on political opinion formation ³⁰.

The Harvard expert survey supplements this finding by documenting divided expert judgement on countermeasures but agreement on their own approaches to health and politics ⁴⁷. This tension between simple manufacturability, concrete case patterns and limited opinion impact should be kept under ongoing observation.

New references have been added to the debate about the discourse itself. Übermedien, AI & Society, Business Punk, TechTarget, Abundance Institute and Al Jazeera's Listening Post classify alarm rhetoric according to the pattern of moral panic or an alarm cycle and demand a shift towards competence, pragmatic management and action ³⁵+538 TechTarget44 Übermedien46 AI & Society (Springer Nature)48 Abundance Institute49 Business Punk. This perspective is broadly distributed in orientation and origin, from Qatari state broadcasting through German media criticism and specialist journalism to libertarian think tanks. The medienrot.de finding on above average trust in public broadcaster AI reporting provides another piece towards the question of where orientation is sought in Germany ⁵⁰.

The thesis about the influence of popular culture images has become more robust through the Terminator contribution ²⁸ alongside the Handelsblatt essay ⁶ and the Frankfurter Hefte classification ³¹ but remains an essay-based thesis rather than a measured effect.

Five axes appear fruitful for further observation. First the industry's handling of its own warnings and the question of regulatory distraction ³+49 MIT Technology Review15 Handelsblatt33 Princeton University39 Yahoo News, second the impact of AI generated false content compared to easy manufacturability ²²+530 Knight First Amendment Institute42 Snopes43 Correctiv45 EU DisinfoLab47 Harvard Kennedy School Misinformation Review, third the role of media as amplifier or corrective ¹⁰+419 Reuters Institute for the Study of Journalism35 Al Jazeera (The Listening Post)44 Übermedien50 medienrot.de, fourth the debate about longtermism and TESCREAL as intellectual background ⁴+320 Reason Magazine32 Asterisk Magazine46 AI & Society (Springer Nature), fifth the German special situation with comparatively high scepticism, particularly in East Germany ³⁶+140 YouGov Deutschland.