AI Lock-In paper warns over-reliance on AI causes deskilling and systemic risk

A new position paper from researchers at (affiliation not stated) argues that AI safety research is missing a critical third dimension: the risk of depending on AI systems themselves. The paper, titled 'Position: AI Lock-In Is in Progress, and We Must Be Prepared' and led by Jaeho Kim with Seokhyun Lee, Jieun Lee, and Changhee Lee, is a spotlight in the ICML 2026 Position Track. The authors coin 'AI Lock-In' to describe a phenomenon where excessive reliance on AI causes human deskilling, diminishes independent functioning, and creates systemic vulnerabilities when AI systems become unavailable or compromised. They argue that while technical alignment and generative AI regulation dominate safety research, this dependence risk remains underexplored and is already emerging at individual, societal, and national levels.

The paper lays out detailed scenarios showing how AI Lock-In escalates — from individual skill atrophy (e.g., professionals losing core competencies because AI handles them) to national-scale infrastructure failures if AI services are disrupted by cyberattacks or geopolitical conflict. The authors stress that these vulnerabilities could be dramatically amplified by service outages or adversarial actions. They provide concrete mitigation guidance at each level, emphasizing that proactive preparation is essential before dependencies become entrenched or irreversible. The work was submitted to arXiv on 28 May 2026 (arXiv:2608.14565) and selected as a Spotlight at ICML 2026, signaling strong community interest. The core message: preserving human autonomy and national security requires treating AI Lock-In as a first-class safety concern, not an afterthought.