The Guardian's headline on Thursday evening: "Safety fears as scientists make first viruses designed by AI." Now read the standfirst printed directly underneath it: "Researchers say breakthrough offers hope for new medicines but also raises urgent biosecurity questions." Two lines, stacked, by the same newspaper, ten minutes of each other's company. The second one is balanced. The first one is a horror trailer. * The science, briefly: a Stanford-led team published work in Science using genome language models — Evo 1 and Evo 2, the biological cousins of the thing that writes your emails — to design complete viral genomes. They synthesised about 300, and 16 of them worked, killing strains of E. coli that had shrugged off naturally occurring viruses. * The word doing the damage is "viruses," alone, unqualified. These are bacteriophages: they infect bacteria and nothing else, they cannot touch a human being, and hospitals already use them to treat patients whose infections don't respond to antibiotics. The Guardian knows this perfectly well — it says so in paragraph two. But a headline travels by itself. On a phone screen, in a push alert, on a share card in someone's group chat, "viruses designed by AI" plus "safety fears" is a lab-leak movie, and paragraph two is not invited. * Be fair, because the fear here is not invented. Researchers at Johns Hopkins have called this an urgent biosafety and biosecurity question, the study's own authors raise it, and the honest worry is about what the same method could do next, in different hands, with a different target. That is a serious story. It is just not the story a reader assembles from the words safety fears and viruses sitting next to each other. * Compare the same paper elsewhere. CNN went long and clunky: "AI creates 16 new viruses from scratch, showing promise for drug resistance and drawing warnings about potential for misuse." Nobody will call that elegant. It is the only one of the three that survives being read alone. The BBC took the opposite route — "Artificial Intelligence used to design brand new viruses" — flat, no alarm and no reassurance, though its third sentence tells you plainly that the 16 viruses "pose no threat to people." * And here is why we're writing this one rather than letting it go. Yesterday we published a piece about the Guardian's previous AI story, the one headlined around a "rogue" system, where the boring explanation — someone had switched off the guardrails — didn't turn up until paragraph 15. Two stories, two days, one desk, and the same shape both times: a careful article wearing an alarmed headline. * Which points at a mechanic rather than a villain. The reporter usually doesn't write the headline. A different person, on a different clock, judged on how many people click, does. That's a normal newsroom, and it explains the gap without excusing it, because the split has a consequence: the sentence with the least reporting behind it is the sentence that gets read the most. * We've spent three pieces this summer mocking outlets for calling AI systems rogue and escaped when the truth was a misconfigured setting. So let's be straight now that the situation has flipped: this time there is a genuine risk worth naming, and the substance of the coverage largely got it right. The bodies were good. The one line that travelled on its own was the one that panicked.