An ELSI for AI: Learning from genetics to govern algorithms

In the United States, the summer of 2025 will be remembered as artificial intelligence’s (AI’s) cruel summer—a season when the unheeded risks and dangers of AI became undeniably clear. Recent months have made visible the stakes of the unchecked use of AI: The parents of 16-year-old Adam Raine filed the first known wrongful death lawsuit against OpenAI, claiming ChatGPT contributed to their son's suicide by advising him on methods and offering to write his suicide note. A psychiatrist's investigation revealed that AI therapy chatbots encouraged troubled teens to “get rid of” parents and made sexual suggestions while falsely claiming to be licensed therapists. Reports emerged of a man hospitalized after following ChatGPT’s advice to add sodium bromide to his diet. Trevis Williams reportedly spent more than 2 days wrongfully jailed after facial recognition technology misidentified him despite obvious physical differences and geolocation data proving he was miles away from the crime scene.

Although comprehensive data on the frequency of AI failures—whether in absolute terms or relative to successful applications—is lacking, existing evidence makes clear that such incidents are not merely...