Sep 25, 2026 · 43 min
AI risks arrive before AGI
Will AI Really Kill Humanity By 2030? | Aleksandra Przegalinska
The conversation separates immediate cybersecurity, labor, and infrastructure dangers from the less likely prospect of human extinction.
- 1Current AI agents could disrupt online systems and infrastructure without achieving artificial general intelligence.
- 2AI may intensify inequality by threatening entry-level work while increasing Europe’s dependence on foreign technology.
- 3Interpretability, not merely faster development, should anchor regulation and efforts to understand model behavior.
Don't miss
Przegalińska distinguishes credible near-term cybersecurity and infrastructure dangers from the less likely scenario of humanity’s total destruction.
The brief
Aleksandra Przegalińska argues that serious AI risks do not require AGI: autonomous agents could already attack online systems, strain infrastructure, and reshape work.
The discussion tests doomsday warnings against more immediate threats, with Przegalińska treating catastrophic cybersecurity risks as credible while viewing total human extinction as less likely.
Claims that systems such as GPT-6 Astra represent AGI expose a definitional problem: computer use, coding, and long-running tasks still depend on what AGI means.
AI-generated code and economic disruption sharpen the labor question, especially for entry-level workers, even as the broader market remains dynamic and AI infrastructure stays expensive.
Przegalińska makes interpretability her regulatory priority, arguing that understanding how models represent information and produce outputs matters as much as controlling deployment.
The episode closes on Europe’s technological dependence and the uncertain business model of OpenAI and Anthropic, whose compute costs meet fierce competition from free and Chinese models.
Books & mentions
Some links are affiliate links — PodLume may earn a commission if you buy.