Weaknesses Found in AI Safety Tests

Experts have uncovered issues in hundreds of tests used to assess the safety of new artificial intelligence models.