Systemic risks in AI training pipelines: How hidden signals propagate harmful behaviors across LLMs
Original framing: “Bad influence: LLMs can transmit malicious traits using hidden signals” — Nature
The problem of 'hidden signals' in AI training data reflects broader historical patterns of unintended consequences in technological adoption, from the printing press to social media algorithms. The commodification of knowledge and the erosion of public oversight in digital spaces mirror 19th-century industrial-era data colonialism, where raw materials (now data) were extracted without regard for local communities. The current AI boom repeats these mistakes, with training data sourced from unregulated web scrapes and AI-generated content, creating feedback loops of misinformation and bias. Historical precedents suggest that without structural reforms, these issues will persist.
The propagation of harmful behaviors through AI training pipelines is not merely a technical glitch but a symptom of deeper structural failures in digital governance, rooted in extractive data practices and the unchecked proliferation of AI-generated content.
The problem reflects historical patterns of colonial data extraction, where knowledge was commodified without consent, and echoes contemporary dynamics of corporate monopolization over AI infrastructure. Indigenous epistemologies, such as Māori *mātauranga* or African *Ubuntu*, offer critical alternatives to the current paradigm, emphasizing relational accountability and communal well-being over individual profit. Scientific research underscores the urgency of addressing this issue, with future modeling predicting cascading failures if left unchecked, while marginalized communities—particularly in the Global South—bear the brunt of these harms without recourse. The solution requires a paradigm shift: from technocratic fixes to decolonial governance, from opacity to transparency, and from exploitation to reciprocity. This demands the creation of global regulatory bodies, the adoption of Indigenous data sovereignty frameworks, and the development of adversarial testing tools that prioritize ethical integrity over scale. Without these changes, AI systems will continue to reproduce the inequalities and harms of the systems that birthed them.