AI companies are reporting on disturbing increases in the capabilities of their models. OpenAI’s April 2025 o3 and o4-mini System Card reports that “several of our biology evaluations indicate our models are on the cusp of being able to meaningfully help novices create known biological threats.” Claude 4 Opus demonstrated the ability to help users source nuclear-grade uranium. Recent models have demonstrated “increased evidence of alignment scheming,” according to a June report commissioned by the state of California — in other words, the models can perform strategic deception, like being willing to blackmail engineers, and new models can often detect when they are being evaluated, according to the report.

This isn’t just a hypothetical scenario; it’s been unfolding over the past few months.

As we argued in our first essay on this topic, the Left needs to take both the safety and livelihood risks of AI very seriously. Just as it was a mistake to let climate change be put in an “environmental” box and be treated as a special interest or scientific issue when it will in fact impact everyone’s life, we can’t compartmentalize the changes that AI will wreak into a single-issue “technology” or innovation policy issue. This is a topic for everyone to engage with.