When AI hallucinations enter parliament, policy-making loses its compass

A sober parliamentary system relies on solid evidence and honest argument. Yet a Guardian analysis has revealed a worrying new threat to that foundation: dozens of policy submissions to parliamentary processes are misrepresenting research, inventing studies, or wrongly citing sources. In many cases the culprit appears to be AI‑generated material — large language models producing content that looks authoritative but does not actually exist. Committee reports have even cited submissions in which a majority of sources appear to be AI‑generated “hallucinations”. These are not mere clerical errors. They strike at the heart of deliberative democracy. Parliamentary committees, ministers and advisers depend on submissions to help understand complex issues, to draw out lived experience and to test policy options. When the raw material of that process is tainted by invented evidence, the quality of scrutiny is degraded and the risk of poor decisions — or decisions based on false premises — rises. A second danger is reputational. When committees cite material that turns out to be fabricated, the public’s trust in institutions takes a hit. Democracy depends on a basic contract: citizens and experts supply reliable information; representatives deliberate and act in the public interest. AI‑generated misinformation corrodes that contract by making it harder to separate genuine contributions from manufactured ones. It is also worth stressing that the system is being “flooded” from across the political spectrum, according to the analysis. This is not a partisan quirk; it is a structural problem arising from easy access to tools that can mass‑produce plausible, but false, academic‑style claims. The technology accelerates the production of content faster than the traditional mechanisms for checking it can keep up. So what should be the response? First, parliamentary secretariats and committees should tighten basic verification procedures. A simple check for the existence of cited studies, journals and author affiliations would catch many of the invented sources described in the analysis. Second, submission portals could require clearer declarations about the use of automated tools and the provenance of material, enabling reviewers to prioritise verification where needed. Third, there should be investment in the skills and resources — both human and technical — that allow committees to detect and flag AI‑generated fabrications. None of this requires rejecting AI wholesale. As with other technologies, there are legitimate uses: drafting, summarising and improving access to information. But those benefits do not absolve us of the responsibility to ensure that what reaches parliamentary desks is true and attributable. For ordinary citizens, the practical implication is unsettling but straightforward: public policy is only as good as the evidence that underpins it. If policymaking is to remain rigorous, transparent and trusted, our democratic institutions must adapt faster than the tools that threaten them. That means better verification, clearer rules on disclosure, and resources to police the quality of submissions. Without those steps, the risk is that decisions affecting everyday life — from health to education, economy to environment — will be shaped by what sounds true rather than what is true.