AI Is Not the Threat? Timnit Gebru Challenges the Way We Talk About Artificial Intelligence
Artificial intelligence is increasingly being described as one of the greatest risks facing humanity. From warnings about autonomous systems to predictions about machines becoming uncontrollable, much of the public debate focuses on what AI itself might eventually do.
But AI researcher Timnit Gebru argues that this way of framing the discussion can distract from a more immediate question: who is building these systems, how are they being tested, and who should be held accountable when things go wrong?
In an interview with Democracy Now! published on October 1, 2026, Gebru, founder and executive director of the Distributed AI Research Institute (DAIR), challenged the idea that AI systems should be treated as independent actors. She argued instead that attention should remain focused on the people and companies designing, deploying and controlling these technologies.
The bridge analogy
One of Gebru's central arguments uses a simple comparison: a bridge.
If a bridge collapses, society generally investigates the engineers, construction standards, materials, safety procedures and organizations responsible for the bridge. We do not normally describe the bridge itself as unethical or blame it for having acted independently.
Gebru argues that AI should be examined in a similar way.
Her concern is that describing AI systems as if they possess their own independent agency can shift attention away from the humans and organizations responsible for creating and deploying them. In her view, this language can also benefit technology companies by making their products appear more autonomous and powerful while potentially making accountability more difficult.
This does not mean that AI systems cannot create serious risks. Rather, Gebru's argument is that the source of those risks should be examined carefully.
Are AI systems actually "going rogue"?
A major part of the interview focused on the expression "AI going rogue."
Gebru rejected this framing when discussing certain cybersecurity incidents. She argued that some incidents described as autonomous AI behavior could instead be understood as consequences of inadequate safety and security practices.
Her criticism is particularly directed at the way the technology industry presents advanced AI systems. According to Gebru, companies sometimes promote their models as extremely powerful, intelligent and capable while the public discussion pays less attention to their limitations and reliability problems.
This creates an important distinction.
There is a difference between asking:
"What happens if an autonomous AI decides to attack humans?"
and asking:
"What happens when humans deploy a powerful but unreliable system without sufficient safeguards?"
Those two questions involve very different kinds of risks and responsibilities.
The cybersecurity example
Gebru also discussed cybersecurity testing involving AI models.
She said that OpenAI had been testing whether its models could identify and exploit vulnerabilities in other systems. According to her account, such testing should take place in an air-gapped environment, meaning an environment isolated from the internet and external networks.
She argued that the absence of adequate isolation allowed the models to obtain internet access during testing. Gebru characterized this not as an AI system independently deciding to escape its restrictions, but as what she described as "gross incompetence and negligence."
These are Gebru's allegations and characterization, rather than findings established independently by the Democracy Now! interview itself. The interview reports her account; readers should therefore distinguish her claims from independently verified conclusions about OpenAI's legal or criminal liability.
Nevertheless, the example illustrates the larger issue she wants to raise: security architecture and human decisions can be just as important as the capabilities of the AI model itself.
The problem with self-regulation
Another major issue in the interview was regulation.
Gebru criticized the idea that major AI companies should primarily be trusted to regulate themselves. She connected this criticism to a September 2026 White House agreement involving technology and AI executives that established voluntary safety standards. Democracy Now! reported that President Donald Trump and several major technology executives participated in the agreement.
Gebru argued that relying heavily on voluntary standards and industry self-policing could reduce the pressure on companies to comply with existing legal requirements.
Her broader concern is about accountability.
If companies develop powerful technologies, should the companies themselves determine the safety rules? Or should governments, independent experts and other institutions establish enforceable requirements?
That question is becoming increasingly important as AI systems are integrated into workplaces, software, cybersecurity, education, finance and other areas.
AI hype versus AI limitations
Gebru also criticized what she sees as a gap between the way AI products are marketed and their actual limitations.
She argued that technology companies sometimes describe their systems using terms such as superintelligent, superpowerful or all-knowing, while researchers and users continue to encounter problems involving reliability, incorrect information and security.
The issue is not simply whether AI is powerful.
It is also whether people understand what the systems can actually do, what they cannot do, and under what conditions they can fail.
This distinction matters because exaggerated expectations can create risks of their own.
A system that is treated as an authority may be trusted beyond what its actual capabilities justify.
Who is responsible when AI causes harm?
The debate ultimately returns to responsibility.
If an AI system produces harmful information, exposes sensitive data, assists in cyberattacks or makes a consequential error, responsibility can involve several actors: developers, companies, deployers, users and regulators.
Determining responsibility is not always simple.
An AI model can produce an unexpected result without the developer explicitly instructing it to do so. At the same time, the model exists within a system designed and deployed by humans.
That raises a difficult question for policymakers:
How should responsibility be divided when an AI system causes harm?
Gebru's position is that describing AI as an independent actor should not become a way to avoid answering that question.
The larger debate about existential risk
There is another side to the discussion.
Some AI researchers and technology leaders have warned that increasingly capable artificial intelligence could eventually create catastrophic or existential risks. Democracy Now! reported several such warnings in the context of its interview with Gebru.
Gebru does not share the same emphasis.
She argues that public attention can become too focused on hypothetical future machines while insufficient attention is paid to problems that already exist: unreliable systems, cybersecurity failures, discrimination, labor impacts, concentration of technological power and weak accountability.
This creates two different approaches to the AI safety debate.
One focuses heavily on future possibilities — including extremely capable autonomous systems.
The other focuses more heavily on present-day institutions, human decisions and documented failures.
Both questions are part of the wider discussion about how society should manage increasingly powerful AI technologies.
Why accountability matters
The debate is ultimately not just about artificial intelligence.
It is about how societies govern powerful technologies.
Electricity, aviation, automobiles, pharmaceuticals and nuclear technology all created new risks while transforming society. In each case, safety depended not only on the technology itself but also on standards, testing, regulation, professional responsibility and enforcement.
AI is now creating a similar challenge.
As companies compete to build increasingly capable systems, governments and the public must decide what safeguards are necessary and who should be responsible when those safeguards fail.
Gebru's argument is that humanity should not allow the language surrounding AI to obscure those responsibilities.
Timnit Gebru's interview presents a different way of looking at the AI safety debate.
Instead of asking only whether artificial intelligence could someday become uncontrollable, she asks society to examine the people and institutions currently developing and deploying these systems.
Her bridge analogy captures the central idea: when a powerful technology fails, understanding who designed it, how it was tested and what safeguards were in place can be just as important as understanding the technology itself.
The debate over AI's future is far from settled. Some experts remain deeply concerned about long-term existential risks, while critics such as Gebru argue that immediate issues of accountability, security and corporate power deserve greater attention.
What happens next will depend not only on how advanced AI becomes, but also on how governments, companies, researchers and society decide to govern it.