The 5 trends to follow in AI according to MIT

Predicting what will happen in AI, in a sector that changes from week to week, is always a risky exercise. Il MIT Technology Review, as happened last year, it will try again for 2026, anticipating some trends that will involve the most talked about sector in the tech world.

In its 2025 report, MIT claims to have identified the success of world models (technologies capable of generating realistic virtual environments in real time), the rise of reasoning models, the boom in AI for science and a closer relationship between AI companies and national security.

In addition to this, he had imagined a competition to Nvidia which, above all because of the TPU on Googlebegan in the last months of the year just ended. From here, however, the list of predictions for 2026 starts.

More Silicon Valley products will be built on Chinese LLMs

For MIT, 2025 marked a turning point for Chinese open source models. The release of DeepSeek R1, an “open-weight” reasoning model, showed that high-end performance can also arrive outside the OpenAI-Anthropic-Google perimeter and with lower entry barriers. The reason is above all operational: open-weight models can be downloaded, run on your own hardware and adapted to your needs.

In this context, the possible growth of Alibaba’s Qwen family is also highlighted, driven by high download numbers and a wide range of variants, as well as other Chinese players who are replicating the same approach. It could happen, therefore, that more Western applications use, often without declaring it, Chinese open models as a basis, while the delay between new releases in China and the Western “frontier” will continue to narrow.

In the United States, the tug-of-war over the regulation of AI will continue

For MIT, 2026 will be another year of clashes over who should set the rules: Washington or individual states. A season of appeals and conflicts is expected, with the largest and most politically equipped states – such as California – more inclined to resist, while others could retreat for fear of pressure or possible repercussions on federal funding.

At the federal level, MIT doesn’t see a single, quick law on the horizon. Instead, we expect a game played above all on lobbying and political power: aggressive campaigns by the industry on the one hand and the growing role of super-PACs on the other. A conflict destined to enter the climate of the midterm elections and, who knows, even to influence them a little.

Chatbots will change the way we shop

The purchasing interface, according to experts, will move completely towards the chat. MIT imagines chatbots increasingly capable of recommending products, comparing alternatives, evaluating strengths and weaknesses, finding the best price and even managing purchases and deliveries. In 2026, according to MIT, partnerships and integrations will increase among the various actors in the field.

An LLM will contribute to an important discovery

The central example is Google DeepMind’s AlphaEvolve: an approach that uses an LLM (Gemini) together with an evolutionary algorithm that checks the proposals, selects the best ones and loops them back to improve them. MIT observes that the path traced by Mountain View is already leading the way: subsequent versions and adaptations are cited (OpenEvolve, SinkaEvolve, AlphaResearch) as well as academic attempts to push reasoning models towards solutions less conservative and more “creative”. With enormous investments in mathematics, optimization, drugs and materials, the prediction is that a significant result will emerge in 2026, not just an incremental one.

Legal battles will intensify

According to the prestigious institute, lawsuits against AI companies will not remain confined to classic disputes over copyright and training data. 2026 should bring more difficult issues to the fore: liability for what a chatbot induces to do, defamation when AI spreads false information about a person, insurance implications for companies in the sector.