As the growth of artificial intelligence accelerates, it is already posing significant challenges to society. Companies have started laying off workers whose jobs AI models can now perform. The expansion of data centres has spiked energy consumption and increased the carbon footprint. AI agents are being used by malicious actors to automate the spread of disinformation. Algorithms are demonstrating bias, raising concerns about flawed application and discrimination by human-related technologies that use them. AI is being used to identify targets in warfare, leading to greater loss of human life.

These are complex problems that demand serious regulatory attention. Yet, today’s dominant AI policy conversations are rarely grounded in these immediate challenges.

Instead, the global discourse has been swept by apocalyptic prophecies. We see this doomsday narrative promoted by some within the AI industry and echoed at the highest levels of decision-making: from lawmakers like Senator Bernie Sanders championing bans on superintelligence, to cautionary messages from the Vatican, to former Secretary of State Hillary Clinton calling open-source AI a real threat.

It is entirely reasonable for global leaders to take these industry-amplified warnings seriously. However, this chorus of influential voices sounding the alarm about AI “existential risk” has inadvertently pulled our focus away from the multiple crises AI has caused, already affecting labour, the environment and communications.

Worse still, the public panic is being weaponised to enforce a modern regime of digital colonisation – the concentration of power in the hands of a few AI corporations in a handful of countries.

The doomsday narrative provides the perfect justification for regulations that would target open-source AI models, which are free or cheaper to use and allow users to modify them to fit their needs. The logic presented to lawmakers is simple and seductive: if advanced AI is a potential weapon of mass destruction, it can only be developed by a trusted few.

Under the banner of “safety”, tech giants are attempting to write the rules to protect their monopolies by shutting down competition from smaller, cheaper open-source AI models.

This directly mirrors historical colonial tactics: preventing local industry from developing to force total reliance on the “empire’s” exports.

Currently, the entire globe is dependent on these frontier AI systems developed by tech giants, especially the Global South, which lacks the resources to create its own AI models from scratch, now costing upwards of $100m in compute power alone.

The problem with using the dominant AI models is not only that they are costly, but also that they often fail in local contexts. Consider Kenya’s recent rollout of the AI-driven Social Health Authority (SHA) system. Because the imported model was built on assumptions that did not accurately capture the reality of Kenya’s vast informal economy, the system failed spectacularly, systematically overestimating incomes and overcharging the poorest citizens for healthcare premiums.

To be effective, frontier AI systems must be adapted to local data. Since building them from scratch is financially impossible for most, open-source AI serves as the only viable bridge for local innovators. More importantly, preserving open-source AI allows the Global South to step off the West’s treadmill of human-mimicking chatbots.

Developing nations do not necessarily need conversational AI systems; they need AI that can help tackle development challenges in areas such as energy, healthcare, agriculture, manufacturing and climate mitigation.

If we crush open-source AI out of a disproportionate fear of AI agents, developing nations lose the main mechanism they have to build their own sovereign tech ecosystems for development.

Furthermore, AI monopolies lock in users in a cycle of data extraction and severe cognitive dependency. We are already witnessing the early stages of cognitive degradation as users increasingly outsource their critical thinking, creativity, and agency to frontier AI systems.

This dependency creates a feedback loop where our growing reliance fuels more data extraction, which in turn builds even more capable systems that strip away human agency and centralise absolute control.

While the Global South is uniquely vulnerable due to historical inequities, this dependency traps anyone not at the frontier. Even wealthy nations in the Global North face the exact same threat of lock-in and extraction.

So what is the solution? Calls for AI bans are certainly not. Abandoning AI would mean letting go of a technology uniquely positioned to accelerate the discovery of new drugs and new solutions to the energy and climate crises. There is already evidence of this with the development of tools such as AlphaFold MOFSeq Banning open-source AI models is also a bad choice. Safety does not negate equitable access to AI. We can, and must, fund robust safety research and implement rigorous testing for near-term breaches. But we cannot allow the fear of hypothetical futures to justify the deliberate consolidation of power in the present.

Safeguarding the future of AI means ensuring that its development is decentralised, transparent, and globally distributed. It means defending open-source innovation as a public good and a fundamental weapon against digital colonisation. It means proactively addressing the human cognitive toll of the technology, ensuring that AI serves to augment human intelligence rather than erode it. It means anchoring absolute accountability in humans, instead of anthropomorphising AI agents that possess no humanity.