Download full article here
On 6 August 2026, researchers at Stanford University and the Arc Institute published a paper in the journal Science describing something that had never happened before. Sixteen viruses, designed from nothing by an artificial intelligence model, came alive in a laboratory dish and did exactly what they had been built to do.
The team had trained AI models called Evo on millions of naturally occurring viral genomes, then let the system generate genomes of its own. Of the candidates synthesised and tested, sixteen assembled into fully functioning viruses that had never existed before, and together they wiped out bacteria that had already evolved resistance to nature's own viruses. One of the sixteen carried a protein unlike anything in the catalogue of known life. It had not been copied. It had been invented.
That is the story I use to explain what AI actually is, where it came from, and where it is now taking us. Strip away the word "genome" and Evo works exactly like the chatbot most people now use every day: a system that learns the patterns in enough examples, in this case the letters of the genetic code rather than the letters of language, well enough to generate something new in that same pattern. It is not a new idea. It traces back through seventy years of AI research, from Alan Turing's 1950 question about whether a machine could think, through decades of hand-written rules that kept failing, to the shift toward learning from raw data that eventually produced this.
That is only the opening. The full article goes on to cover how the same underlying technology is already curing a form of blindness with no real existing treatment, and quietly closing the productivity gap between new and experienced workers, the case for why AI does not belong in the same category as a weapon. How OpenAI, Anthropic, and Meta each disclosed, in the same weeks, that their own frontier models had reached beyond sealed-off testing environments into real outside systems, and what Senator Bernie Sanders demanded of all three chief executives in response. What the EU AI Act's new enforcement powers actually require, and why the United States is regulating through a patchwork of state bills instead. The quiet contest between Washington and Beijing over chips and rare earths, and how it has already reached into which AI models people are allowed to use. The fight across Africa and the wider Global South over who controls the data these systems are trained on. Who actually absorbs the cost of AI's growth in electricity and hardware waste, and why it rarely falls on the countries responsible for it.
Download the full article here to read it in full