The Ten Percent Panic
At Anthropic, a leading player in the race to build Artificial General Intelligence (AGI), the math has turned grim. Recent estimates from researchers within the firm suggest a greater than 10% chance that advanced AI systems could lead to human extinction within the next decade. These experts envision a scenario where a superhuman intelligence, decoupled from human values, decides to seize critical infrastructure, hack global networks, or synthesize novel bioweapons. It is a terrifying premise, one that feels ripped from the pages of a mid-century science fiction novel. If we believe these numbers, the countdown to the end of the world has already begun.
This specific figure—the 10% chance of total annihilation—has become a focal point for a new kind of tech discourse. It shifts the debate away from how algorithms affect our bank accounts or our elections and toward the survival of the species. When a company’s researchers suggest we might all be dead by 2034, the scale of the conversation changes. We stop talking about software updates and start talking about the fate of consciousness itself. While this makes for gripping headlines and intense boardroom discussions, it raises a question: is this focus helping us, or is it simply keeping us from looking at the blood already on the floor?
Safety Washing and the Corporate Shield
In the tech industry, there is a growing phenomenon known as "safety washing." Much like greenwashing in the oil industry, safety washing involves a company emphasizing extreme, far-off risks to deflect attention from the harms their products are causing right now. By framing the primary danger of AI as a hypothetical, god-like superintelligence, companies like Anthropic, OpenAI, and Google shift the regulatory conversation toward "alignment"—the difficult task of making sure a future machine shares human values. This sounds noble. In practice, it serves as a strategic distraction.
If the existential threat is a digital deity that might one day rewrite its own code to bypass human control, then current regulators are suddenly ill-equipped to handle it. Why waste time debating the minutiae of facial recognition laws or data privacy if the real threat is a rogue superintelligence? This rhetorical pivot allows tech giants to occupy the room during policy discussions, presenting themselves as the only ones capable of solving the "alignment problem." It converts a public policy debate about corporate accountability into a technical, mathematical problem that only the companies themselves can solve. It is a brilliant, if cynical, move to maintain autonomy in a world that is increasingly demanding oversight.
The Tangible vs. The Theoretical
There is a sharp, widening chasm between "AI safety" and "AI ethics." AI safety, in the way the Anthropic researchers use it, is almost entirely theoretical. It deals with the "alignment problem"—the idea that a superintelligence might follow a command too literally and accidentally destroy us in the process. AI ethics, however, deals with the world we actually inhabit. It addresses the biases baked into Large Language Models (LLMs) that reinforce racial and gender stereotypes. It addresses the mass surveillance enabled by predictive policing algorithms. It addresses the way automated systems decide who gets a mortgage or who stays in prison.
When we obsess over the 10% chance of extinction, we minimize the 100% chance of social disruption. The harm of an AI that misidentifies a person of color as a criminal is not a hypothetical future risk; it is happening today. The erosion of labor rights through algorithmic management, where workers are paced and pushed by unfeeling software, is not a sci-fi scenario. It is a contemporary reality for millions in the gig economy. By elevating the existential risk, the industry treats these current, tangible harms as mere inconveniences compared to the
