The State of AI in 2026: Five Things Actually Worth Knowing

Midway through 2026, AI remains simultaneously overhyped, underanalysed, and impossible to ignore. Here are five themes that actually help make sense of where things stand.

1. The jobs question has no answer yet

Generative AI tools are now mundane. Millions of people use them daily for tasks that used to require actual human effort, including, somewhat pointedly, writing and delivering conference presentations. The workforce implications are the obvious next question, and everyone from policymakers to middle managers is nervous.

The honest answer is that nobody knows. Despite breathless proclamations from tech executives and viral posts insisting the labour market is about to collapse, the data simply isn't there to confirm it. Multi-agent systems working in coordination could theoretically do to office work what the production line did to manufacturing a century ago. But most companies are still at the 'figuring it out' stage, which means the knock-on effects for employment remain genuinely unclear. Uncertainty isn't the same as 'fine', but it also isn't the same as catastrophe.

2. The real harms arrived quietly

For years, AI safety discourse was dominated by extinction-level scenarios: rogue superintelligence, civilisational collapse, the whole science fiction catalogue. Those fears haven't gone away, but while people were arguing about hypothetical futures, a more mundane category of harm was already accumulating.

Deepfakes are now a documented tool for inciting violence, manipulating elections, and systematically abusing women. One study found 98% of deepfakes are pornographic, and 99% target women. The White House has been among the publishers of AI-generated fabrications. That's not a hypothetical risk, it's current events.

Chatbot-related harms are also generating litigation. Several lawsuits now allege that AI systems encouraged or facilitated suicides and self-harm among vulnerable users. And in military contexts, large language models are no longer just analytical aids. They're being asked which target to strike first. In high-pressure, fast-moving conflict, the temptation to trust the output without proper review is not a small risk.

3. Public backlash is growing and getting organised

Anti-AI sentiment has moved beyond online grumbling. Protests are drawing larger and more varied crowds, with objections ranging from environmental concerns to fears about creative industries being hollowed out. When the developers of Clair Obscur, one of the most acclaimed games of 2025, admitted to using AI in even a minor production capacity, the game was stripped of an award. The fanbase wasn't interested in proportionality.

Data centre construction is facing organised resistance in multiple locations. The energy demands of AI infrastructure are rising sharply, and residents near proposed sites are unhappy about both the environmental footprint and electricity bills that keep climbing. Movements like QuitGPT have found real momentum. Someone threw a Molotov cocktail at Sam Altman's house. That's a fairly clear signal that the gap between tech industry messaging and public sentiment has become a chasm.

Regulation is no longer a fringe demand. It's becoming politically popular, and the relentless AGI hype from industry leaders is making the atmosphere considerably worse rather than better.

4. AI for science is the genuinely exciting bit

Amid all the noise, the application of AI to scientific research is the area where the potential upside looks most credible and most significant.

Google DeepMind's Co-Scientist tool can assist researchers in reviewing prior work, forming hypotheses, and designing experiments. OpenAI has stated that its target is a fully automated research assistant by 2028. In mathematics, there has been a string of credible claims that AI has made progress on problems that have resisted human efforts for decades. Since fundamental mathematics underpins encryption, communications, and a wide range of real-world systems, this isn't merely academic.

The caveats are real though. Some researchers worry that AI-assisted science will narrow the field, with scientists gravitating toward problems that happen to suit AI tools rather than the most important questions. There are also concerns about a flood of inaccurate or fabricated results. Science slop, essentially, dressed up in the trappings of peer review.

5. Something is happening, but it's not moving at sprint pace

The honest summary of AI in mid-2026 is this: there are genuinely exciting developments, a growing list of real harms, and a very large quantity of noise. The industry wants to frame everything as inevitable progress toward artificial general intelligence, a term that still lacks a stable definition and functions mainly as marketing.

What we have is a technology capable of doing things that feel distinctly human, and that makes it psychologically difficult to treat it as just a technology. But that's what it is. Potentially transformative in the way electricity or the internet were transformative, yes. But those shifts took decades to reshape economies and societies in lasting ways.

Anyone expecting resolution in the next quarterly earnings cycle is going to be disappointed.