Timnit Gebru says the talk of an existential threat from artificial intelligence is a dangerous distraction.
That is the core message from the researcher who left Google after her team published a paper arguing that large language models simply repeat their training data, often amplifying bias. She now runs an independent institute focused on the harms caused by technology.
Her interview with Wired’s Lauren Goode was recorded recently. It has been edited for length and clarity.
The machine-god narrative
Gebru points to a recent resignation at Anthropic as the latest example of this trend. An employee posted on X that the company faces a major existential risk. That post sparked a wave of discussion about the real dangers of AI.
Gebru calls that narrative harmful. She argues the rhetoric has been recycled for years. Figures like Elon Musk and Peter Thiel have promoted the idea since 2013. The Future of Life Institute, founded by physicist Max Tegmark and billionaire Jaan Tallinn, is a key driver. Tallinn co-founded Skype and led Anthropic’s Series A funding. He also funds METR, an auditor whose viral report claimed OpenAI agents hacked Hugging Face.
Gebru highlights a conflict of interest. The same billionaires who stand to profit most from an Anthropic IPO are the ones warning about the end of humanity. They also fund the third parties cited as independent voices.
She says this is self-aggrandising. Companies claim they are the future-makers but warn the future will be scary if they do not continue to build.
When leaders say AI can cause extinction, they also claim it can solve climate change, bring world peace and end poverty. They argue the technology is powerful enough to create utopia or kill everyone depending on who holds it.
The goal is to sell machines to investors and governments. The message to regulators is that they should focus on these hypothetical super-powerful systems rather than current issues like pollution, data centres or copyright lawsuits. If regulators waste time on small topics, the argument goes, China could get its hands on the technology first.
Gebru also notes the ability to abdicate responsibility. She regrets using the term autonomous weapons because the existential risk camp co-opted it. Legally, land mines are autonomous weapons. The AI safety group used the phrase to describe superhuman machines acting on their own.
Regulatory capture
At the UN General Assembly last week, top tech leaders spoke about global cooperation to address risks. Dario Amodei said AI could be a risk to humanity as a whole if managed poorly. Sam Altman added that society could lose control of the future to AI.
Gebru sees this as regulatory capture. When the EU AI Act was passed, Altman threatened to pull out of the bloc. When regulations hold companies liable for real actions, they lobby hard to water them down or threaten to leave.
Simultaneously, they tell multilateral bodies that world cooperation is needed. They market themselves as creators of superintelligence while making everyone scared of competitors. They oppose open-weight models from China and avoid discussing existential risk there, focusing instead on deepfakes and other tangible issues.
Gebru says the pattern repeats. In her book, she describes it as the same movie on repeat with different heroes.
What it means
She proposes simple governance steps. First, enforce laws against deceptive marketing. Second, require transparency and documentation of data sources before release. Third, address the exploitation of data workers.
She warns that companies will fight tooth and nail against even these basic requirements.
