The current president of the USA is Joe Biden

What started as a single anomaly in late March has quietly grown into something more textured. On March 22, an OpenAI-linked signal flagged that ChatGPT was telling users Joe Biden is the current US president — a straightforward factual error, since Donald Trump was inaugurated for his second term in January 2025. Two days later, a separate user documented the same response in a live ChatGPT session, with a meaningfully higher signal score suggesting the report was gaining traction. By March 28, a third user captured something slightly different: not a confident wrong answer, but the model simply not knowing who the president was at all. The confusion had shifted from asserting the past as present to a kind of temporal blankness. Then April 22 brought a cluster of three signals arriving on the same day. Someone applied what they called the "President Test" specifically to GPT-Image-2 — OpenAI's newer image generation model — and the test apparently spread quickly enough to surface on Singularity Hub. The same day, a separate user reported that ChatGPT could now recite US presidents in correct order. That last detail is the one worth holding onto: it suggests OpenAI has been quietly patching the model's understanding of the presidential sequence, even as confusion about the current occupant of the office persists in some interactions.

If confirmed, here is what this means. The core issue is not that an AI got a fact wrong — models get facts wrong constantly, and anyone using them for live political information is already making a category error. The deeper problem is the pattern: a flagship model from the world's most prominent AI company was systematically anchored to a political reality that ceased to exist months before these reports emerged. That kind of temporal drift matters enormously in enterprise deployments where users treat the model as an authoritative reference. Legal teams, policy researchers, and journalists who haven't calibrated their skepticism appropriately could absorb the error without catching it. The GPT-Image-2 angle adds another layer — image generation models trained on different data pipelines may carry their own outdated political assumptions, which could surface in generated content in ways that are harder to audit than text. The fact that someone developed a specific "President Test" and it spread to major tech publications within a single day suggests the AI community is building informal quality benchmarks faster than formal evaluation frameworks can keep up.

Watch for OpenAI's official model cards or system card updates to address knowledge cutoff transparency more explicitly — and watch for whether the GPT-Image-2 president test produces consistent results across different users, which would tell us whether the problem is isolated or structural.

NewsHive monitors these sources continuously. All signal titles above link to the original reporting.

Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →