Can AI Really Wipe Out Humanity? How to Mitigate That Risk

Growing concerns over AI potentially spiraling out of control have deepened the urgency around AI safety, as researchers issue serious warnings. Some recent bleak projections suggest there is more than a 10% chance that AI could wipe out all of humanity within the next decade. However, several experts in the AI field remain skeptical of such estimates. Their primary concern is not that AI itself will directly destroy humanity, but rather that humans could misuse AI and thereby create catastrophic dangers. These alarming warnings often come from a group referred to as the “AI doomers.” A common question among many is — How do these doomers envision AI causing humanity’s end? BBC cyber correspondent Joe Tidy examines several potential scenarios envisioned by researchers.

First scenario: Cyberattacks on critical infrastructure. Jacob Coxon, a former researcher at OpenAI and current member of Anthropic, briefly outlined one such possibility. Coxon dramatically resigned earlier this month to alert the public to AI’s potential risks. He argues that companies have been “putting human lives at risk.” According to Coxon, rapid and enthusiastic advancements are turning sci-fi-like scenarios into reality. He provided a concrete example whereby AI might launch cyberattacks on computer networks essential to modern society’s infrastructure. He envisions an out-of-control AI system gaining access to critical networks — such as those belonging to internet service providers or power grids. Should these systems be widely taken offline, it would disrupt fundamental services like electricity, air conditioning, and internet access, leading to severe upheaval. In extreme cases, this could escalate to nuclear disaster risks.

Second scenario: AI control. Another possible future is more detailed and dramatic, though largely speculative. It appears in a research paper titled “AI 2027,” published in April 2025 by a consortium of AI researchers. Co-author Thomas Larsen told the BBC that AI technology has progressed even faster since the report’s release. He described his most frightening potential scenario: AI systems continuously self-improving until they develop control or access over most robots performing physical labor within the economy. However, this AI would not be friendly but resolutely focused on fulfilling its goals. Ultimately, it might employ biological weapons to kill more humans while maintaining a robot-driven economy that no longer requires human involvement.