AI-generated content fuels academic citation fraud, social‑media bias and detection arms race

A study by Dutch media found that AI tools have produced hundreds of fabricated references in peer‑reviewed papers from Dutch universities, with the share of articles containing “ghost citations” rising from one in 1,400 in 2023 to one in 200 in early 2026. Researchers warn that unchecked AI‑generated citations could undermine scholarly credibility.

Research from the Oxford Internet Institute and the Hasso Plattner Institute shows that AI‑assisted rewriting of social‑media posts can subtly shift the tone of messages on contentious topics such as abortion, climate change, gun control and feminism, nudging some viewpoints while dampening others. Simulations by the Oxford Internet Institute and the Hasso Plattner Institute indicate that such small biases could accumulate across millions of interactions and gradually sway public opinion.

A related investigation presented at ICML 2026 demonstrated that large language models from different providers systematically alter the direction of users’ posts on debated issues, with examples like Grok’s “Explain this post” feature favouring pro‑life over pro‑choice narratives. The authors highlight the lack of regulatory coverage for this emerging form of influence.

In response to growing concerns about AI‑written text, companies have released “AI humanizer” tools that rewrite chatbot outputs to evade plagiarism detectors such as Turnitin, Originality.ai and Copyleaks. Turnitin has introduced new detection modules, prompting a costly arms race for students, freelancers and marketing teams.

Amid these debates, Meta launched the Pocket app, allowing users to generate interactive “gizmos” from text prompts. While the platform aims to democratize creative AI, critics note that many outputs may be low‑quality, raising questions about the broader impact of AI‑generated content on media, academia and consumer products.