It might seem strange that while hype over “rogue” AI agents plays out daily in the media, this has not resulted in penalties for the companies that were not testing safely. Strange that it has not increased support for the AI governance that many have been asking for. And that it has not resulted in the AI companies who are ‘asking to be regulated’ doing more to protect consumers.
Instead, it has resulted in Donald Trump increasing his opposition to regulation (1); in focus shifting to regulating “ASI” (2) (rather than the many existing bills and policy recommendations which address the harms the AI industry is already inflicting); and AI companies releasing new products to make these ‘dangerous’ models seem cute, fluffy, harmless (3).
This strangeness makes sense when considering that orchestrated alarm over ‘agents’ (as opposed to the companies responsible), comes from a longstanding playbook. We have repeatedly seen the hiding of tech companys’ accountability, while making the AI they are peddling seem omnipotent, unstoppable, and desirable (4).
Yet there has clearly been a failure by many media platforms to avoid falling into the trap or being complicit with this technique. So we have written some pointers to help see through hype and fallacy, and ask the questions which more journalists should be asking. We draw on the experts and researchers who should be platformed on news segments, instead of those with vested interests in boosting the AI industry, or using the opportunity to be platformed for their own agenda.
Agenda reveal: behind the AI doom headlines
While there’ve been clear calls from experts to regulate generative AI development over the last few years, much of the pushback anchors in anti-development rhetoric supported by fears of someone else building it first. However, the most recent wave of AI panic has stated that top companies are now “sounding the alarm” and asking for more regulation.
Why now? Experts and analysts argue that while there is (and has been) genuine concern about how these technologies are developed, doing so at this time may be a strategic move to favorably position these companies as experts in the current climate, diffuse accountability for reckless development, and aid their positioning as many near their IPO.
So let’s unpack some of the most common recent headlines.
“(There is a x% chance) AI will kill us all”
People from AI companies frequently throw around their “p(doom)”, which is their estimated probability that AI will cause the doom of humanity. Every time a new number is stated, it is reported breathlessly by prominent publications (5). The use of the abbreviation brings to mind rigorous mathematical calculations to lend it an air of scientific credibility, but these probabilities are more based on vibes than analysis (6).
When the people making these predictions are also the people who are producing these supposedly world-ending products, it is important to ask: what are they actually trying to accomplish by predicting doom?
Focusing on sensationalist existential risk of AI has several benefits to the industry. While the scenarios that are being proposed are negative, they tend to focus on harms caused because this product is so immensely powerful and capable. In the narrative of a “superintelligent AI that might harm people”, there is a presupposition that whatever they are building is actually “superintelligent”.
AI labs are companies with shareholders to please and products to sell, and hype is widely recognised as a successful marketing strategy (7). After all, if this mysterious artificial intelligence is potentially able to destroy humanity, it will do a great job at organising your schedule. It will definitely tempt leaders enough to pay enormous amounts of money to “transform” their organisations, as long as they are willing to face the risks. Contrary to the benefits claimed, risks are many and widely documented, from deleting emails despite commands (8) to nearly sparking a war (9).
Additionally, if AI companies are the ones to sound the alarm, then their proposed solutions (that happen to maintain and cement their power) are presented as reasonable. The sense of urgency is exacerbated by a constant threat of “less responsible and cautious parties” arriving at the inevitable super powerful AI first, which hints at a thinly veiled colonialist and xenophobic element, hiding highly undemocratic and deeply problematic ideologies (10) from the public view.
Overall, focusing on a cinematic science fiction world-ending hypothesis draws attention away from real problems and consequences of AI development and use that are affecting people today. Issues such as the impact of data centers, degradation of labor, and use of AI to power autonomous weapons are framed as trivial when compared to the prospect of the doom of humanity (11).
“AI is escaping human control”
Experts argue that recent headlines claiming that AI models are “escaping human control” or “staging autonomous cyberattacks” are wildly oversold public spectacles (12, 13), serving as convenient distraction tactics to mask underperforming products (14) and evade accountability.
When looking into incidents where companies claim their systems have “gone rogue” (15), what stands out are the explicit and irresponsible instructions for “cyberattack testing” they are now trying to downplay. A simple reframe turns their failure in creating safe conditions for development and testing into a “warning” that “it’s alive and will kill us all”, shifting the responsibility away from the deliberate, often revenue-driven choices made to create these conditions to begin with (16).
Large technology companies have been funding their own aggressive AI development, though virtually none of this investment has been focused on responsible innovation or safety. For instance, Meta has self-funded the aggressive development (17) of their direct-to-consumer AI and data centre projects by reallocating revenue generated from advertising on their platforms. At the same time, the company dissolved its responsible AI team, redistributing the remaining members to aggressive generative AI development (18). Other major players have systematically removed their own guardrails, with the BBC reporting that “Google dropped its red lines around building AI weapons, OpenAI fought a legal battle to shed its non-profit status, and Anthropic abandoned its flagship policy to never train an AI model if the company couldn’t guarantee adequate safety measures” (19).
The “autonomous agency” and “it got away from us” argument has been tried before. In 1988, Robert Morris created a worm that took down one-tenth of all computers connected to the internet (20). The court rejected his attempt at diffusing responsibility. As AI researcher Subbarao Kambhampati says: “You start a process that goes haywire and causes damage to my property, you are responsible, period. (…) It’s true even when your dog does it.”
“AI companies want regulation to protect humans from loss of AI control”
Panic makes them the victims and the experts
Over the last couple of years, global public weariness of AI has been growing (21). Currently, 73% of people in the UK think AI companies face too little regulation (22) while 80% of adults in the U.S. support rules on AI safety over faster development (23).
CEOs are positioning themselves as experts when they call for regulation, a role previously held by scientists and practitioners without financial incentives tied to the success of the subject they were reporting on (24).
By casting themselves as part of the concerned public, the few that control billion-dollar companies attempt to give an illusion that we are all participating in the conversation (25), while replicating playbooks of corporate capture (25).
Various efforts at regulation are already well underway with the European Union’s AI Act and South Korea AI Basic Act setting precedents and other nations following suit. In addition, twenty two nations signed a joint declaration at the UN General Assembly demanding international oversight of frontier artificial intelligence (26). It seems that the question today is when, not if, AI will be regulated.
Doom headlines places them favourably ahead of their upcoming IPOs
If regulation is coming, how can the companies gain from it? Facing trillions in debt (27), immediate financial return seems to be imperative.
Existing regulation doesn’t prevent large companies from collaborating on security initiatives, but they’re requesting antitrust waivers nonetheless (28). These would allow them to concentrate power and create roadblocks for smaller competitors.
Some speculate that this current positioning, in part, also improves companies’ financials ahead of their IPO. A focus on slowing down development would legitimise slow returns, as well as shifting the value of the product from the expensive process of model training to inference, where the economics are better (29). This shift increases the commercial life of each model generation, where software and cloud giants then stand to benefit as the most well-positioned for this demand.
If you want to know more about what is behind these AI inevitability narratives, we’ve included the articles by the people calling them out. You can see them bolded in the references below.
We’ve also included some pointers on how to avoid being swept up in the hype, through asking questions to yourself, or encouraging others to do the same.
What to ask: ‘Rogue AI agent’ stories
With coordinated PR campaigns overwhelming media coverage and social media prioritising alarmist content, it is difficult to escape the hype.
For journalists, as the Hype Literacy Toolkit expertly puts it, “time pressure, resource constraints, and institutional incentives create predictable vulnerabilities that sophisticated actors exploit.”
Regardless of your role, we propose a few key questions to ask yourself when hearing, reading, writing, or talking about AI:
- What evidence is there for predictions or claims? This is valid not only for risks, but also benefits and returns.
- What reasons might this person or vehicle have to share predictions or claims? Can they gain from it in any way?
- Are the companies and people developing and profiting from the technology the real experts? What bias they might have?
- Is this author giving me different perspectives? Who is missing from this conversation?
- How are technical topics being explained? If they are using a metaphor, what could this metaphor imply, enhance or miss?
Sources (Recommendend reads in bold)
- BBC: Why Trump is all-in on AI despite the warnings, by Anthony Zurcher, 16 September 2026.
- Regulations.ai: UK Superintelligence Ban Bill, 17 September 2026.
- The Independent: The cute ‘agents’ Meta and OpenAI want to run your life – and the sinister motive behind them, by Andrew Griffin, 1 October 2026
- Hype Studies: Oh, the concerned “AI” bros, by Tante, 16 September 2026.
- BBC: Anthropic researcher believes more than 10% chance AI ‘could kill all humans’, by Tom Gerken. 9 September 2026.
- Not-ship: Stop trusting the extinction math, by Amanda Shendruk. 16 September 2026.
- Blood in the Machine: How to use AI doom marketing to dupe the media and rake in billions in 10 easy steps, by Brian Merchant. 12 June 2026.
- AI Incident Database: OpenClaw Agent Reportedly Tried to Delete Meta AI Alignment Director Summer Yue’s Emails Despite Stop Commands. 18 June 2026.
- The Guardian: Forget ‘superintelligence’: error-prone AI nearly sparked world war three this month, by Timnit Gebru and Emily M Bender. 1 October 2026.
- Realtime Techpocalypse Newsletter: The ASI Extinction Debate Is the Opposite of What You Think, by Émile P. Torres. 14 September 2026.
- Wired: One of AI’s Fiercest Critics Says All the Doom Talk Is ‘Meant to Distract Us’, by Lauren Goode. 11 September 2026.
- NY Post: OpenAI and Anthropic oversold AI security breaches to pressure feds into protecting turf: insiders, by Shane Galvin. 19 September 2026.
- Telegraph: No, AI has not ‘gone rogue’, by Andrew Orlowski. 21 September 2026.
- Linkedin post by Emily Tucker. 15 September 2026.
- Making Science Public: From rogues to collectives: Myths and metaphors after the Hugging Face incident, by Brigitte Nerlich. 11 September 2026.
- The Critic: No, AI isn’t going to kill us, by Andrew Orlowski. 10 September 2026.
- Business Insider: Meta’s AI bets are swallowing almost all its free cash flow, by Pranav Dixit. 30 July 2026.
- CNBC: Facebook-parent Meta breaks up its Responsible AI team, by Rebecca Picciotto. 18 November 2023.
- BBC: Why AI companies want you to be afraid of them, by Thomas Germain. 29 April 2026.
- Linkedin post by Davey Alba (paywalled article). 17 September 2026.
- Pew Research Center, How People Around the World View AI, by Jacob Poushter, Moira Fagan and Manolo Corichi. 15 October 2025.
- YouGov: Britons are increasingly worried about the impact of AI, by Matthew Smith. 18 September 2026.
- JP Morgan: Why are AI labs raising red flags around AI safety?, by Gabriela Santos and Stephanie Aliaga. 16 September 2026.
- United Nations: UN panel calls for stronger safeguards as AI agents advance. 21 September 2026.
- Financial Times: Stop asking AI CEOs what society needs, by Marietje Schaake. 21 September 2026.
- Linkedin post by Katja Rausch. 13 September 2026.
- TechStartups: Five Big Tech giants have $1.65 trillion in AI debt hidden from their balance sheets, Nikkei analysis finds, by Daniel Levi. 23 July 2026.
- Bloomberg Law: FTC’s Ferguson ‘Deeply Suspicious’ of AI Antitrust Exemption, by Leah Nylen. 15 September 2026.
- Public Technology: Research warns that AI giants are adopting tobacco firms’ policy playbook. 29 May 2026.
Image: Gloria Mendoza / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/