# Gebru: AI existential risk talk is a dangerous distraction > Timnit Gebru argues that warnings of AI destroying humanity are marketing tactics by billionaires who profit from the technology, not genuine safety concerns. - Title: Gebru: AI existential risk talk is a dangerous distraction - Summary: Timnit Gebru argues that warnings of AI destroying humanity are marketing tactics by billionaires who profit from the technology, not genuine safety… - Keywords: timnit gebru, ai safety, existential risk, stochastic parrots, large language models, regulatory capture, ai ethics, anthropic, openai, effective altruism, Gebru, existential - Source: WIRED — https://wired.com/story/the-big-interview-podcast-timnit-gebru - Author: Lauren Goode - Published: 2026-09-29T10:30:00+00:00 - Read time: 4 min - Topics: timnit gebru, ai safety, existential risk, stochastic parrots, large language models, regulatory capture, ai ethics, anthropic, openai, effective altruism ## The new AI vocabulary As artificial intelligence booms, new terms have emerged in debates over its consequences: stochastic parrots and existential risk. Both centre on researcher Timnit Gebru. ## Gebru's Google fight Gebru coauthored a paper calling out biases in Google's AI, saying large language models parrot training data and risk perpetuating bias. Her departure led her to found an institute investigating technology harms. ## The existential risk debate Gebru has spoken against industry figures who believe AI could destroy humanity. Some, including an Anthropic cofounder, have alleged her earlier research on stochastic parrots is no longer relevant. ## A long history of warnings Elon Musk and Peter Thiel have said AI poses existential risk since two thousand thirteen. The same rhetoric repeats every three years from prominent figures. ## The same billionaires, multiple roles Max Tegmark and Jaan Tallinn founded the Future of Life Institute warning of AI existential risk. Tallinn, a Skype cofounder, also led Anthropic's Series A funding and funds METR, a so-called independent auditor. ## Gebru's core claim The same billionaires who founded and funded Anthropic, who stand to benefit most from its IPO, also founded institutions warning of AI existential risk and funded third parties cited as independent. ## Why the narrative is harmful Gebru calls the existential risk narrative harmful, not just a distraction. It is dangerous because it serves the financial interests of those promoting it. > It's not just a distraction, but it's harmful. ## The counterintuitive pitch If billionaires stand to profit from AI company IPOs, why do they say those companies might build something that kills us all? Gebru says it sounds counterintuitive but serves their interests. ## Self-aggrandisement and power By claiming AI is so powerful it could destroy humanity, companies position themselves as future-makers. They tell investors, governments and regulators they need access to super powerful machines. ## Distracting from real harms When companies talk of existential risk, pollution and data centre impacts sound small. Copyright lawsuits and actual harms fade from regulatory focus compared to fictional super powerful machines. ## Abdicating responsibility The existential risk narrative allows companies to abdicate responsibility for real harms. When agents go rogue or systems fail, blame shifts away from engineers who built the tools. ## UN stage, same message At the UN General Assembly, Sam Altman said we could lose control of the future to AI. Dario Amodei reportedly said AI could be a risk to humanity as a whole if managed poorly. > If managed poorly, I even believe AI could be a risk to humanity as a whole. ## Regulatory capture in action Gebru calls this regulatory capture. When real regulation threatens, companies threaten to pull out or lobby hard. In two thousand twenty-three, Sam Altman threatened to leave the EU after it regulated AI. ## What stochastic parrots means Large language models are trained on vast internet text to output the most likely text sequences. They power chatbots like Claude and ChatGPT. Stochastic parrots is a metaphor for this pattern-matching without understanding. ## Jack Clark's attack Anthropic cofounder Jack Clark posted that stochastic parrots was a memetically fit cognitive virus from two thousand twenty-one to two thousand twenty-five that blinded people to AI progress. ## Gebru's rebuttal Gebru says the research is not outdated. Large language models are still large language models. What they are has not changed and will never change. The definition remains the same. > Large language models are large language models. Now, you might have large language models in a separate system. They now have reinforcement learning agents. But our paper was about large language models, and that's never changed. ## Real harms from pattern matching Stochastic parrots cannot understand text content, so they cannot be factual. A Palestinian man's good morning was translated as attack them. Over-trust in automated systems causes real medical misdiagnosis. ## No reasoning, just patterns Gebru rejects claims of reasoning or recursive intelligence in AI. These are aspirational names. When benchmarks change slightly, reasoning claims break down, showing systems rely on patterns, not understanding. > There is no reasoning. There is no recursive intelligence. ## Demanding real science Gebru says real evaluation takes time. Companies should provide training data, evaluation data and methodology so researchers can reproduce results. Press releases should not become law without scrutiny. ## Gebru's hope and focus Gebru finds hope in people resisting through media, narrative, data centres and funding. She believes in human agency and collective power to imagine better futures and stop bad things from happening. ## Key takeaway AI researcher Timnit Gebru says existential risk narratives are harmful marketing by industry leaders to distract from real harms like labour exploitation and data theft.