Frontier AI may be hard to specify, but not impossible to govern: this piece argues for measurable safety thresholds, release gates, and accountability.