Ten forces shaping AI in 2026: humanoid training data, war-room tools, and the deepfake threat

AI’s once-singular storyline has splintered into a set of powerful, competing currents. A snapshot of the field in 2026 points to ten developments that will steer what gets built next—and how the technology collides with everyday life in the United States and beyond.

Humanoid robots are learning from us at scale. Vast troves of video capturing human movement are being fed into training pipelines, including footage gathered in sprawling “training centers” where workers repeat tasks for the cameras and from tele-operated machines effectively “puppeted” by people overseas.

It’s an ambitious, sometimes bizarre effort—and one with no guarantee of success. Large language models may have lost their novelty, but they have not run out of road. While researchers chase the “next big thing,” the prevailing view is that there is still significant performance and utility to extract from LLMs, even as obvious gains become harder to find.

Security risks are escalating. AI is lowering the barriers for scammers and hackers, making attempts to infiltrate targets faster, cheaper, and easier. At the same time, in the U.S., LLMs could supercharge mass surveillance by making commercially available bulk datasets even more of a privacy concern.

Companies are racing to build systems that understand the external world. If they succeed, such models may overcome limitations of text-dominated LLMs and help AI operate reliably in physical environments—from factories to homes. In the military sphere, generative AI now has its own seat in the war room.

Algorithms that once handled logistics and analysis have been joined by systems whose advice commanders take seriously, reshaping how militaries share intelligence, work with Big Tech, and make lethal decisions. The long-predicted threat of weaponized deepfakes has arrived.

Between improvements in generative models, Grok’s mass generation of nonconsensual sexual images, and a U.S. administration using the technology for propaganda, policymakers and platforms are confronting a more immediate and volatile information environment. AI agents are evolving from solo operators into cooperative teams.

Early agents could run a browser or write code snippets on their own; the next wave coordinates multiple agents to tackle far more complex, multistep goals. Openly releasing cutting-edge models has become a geopolitical strategy. By giving away frontier systems for free, Chinese labs have won global credibility and developer goodwill.

Whether that approach is financially sustainable is unclear—but the world is already building on Chinese foundations. Inside labs and universities, researchers are developing AI co-scientists: agents that can autonomously carry out research tasks and collaborate with humans.

Some believe these systems could eventually reach achievements on par with Nobel Prize–worthy work. Public sentiment is shifting from fascination to fatigue—and pushback. After years of unfettered development, a backlash is gathering strength across the political spectrum and among artists and labor unions, with activists gaining momentum and notching small wins.

AI is everywhere, all at once, and that ubiquity is forcing a harder conversation about what progress should look like. Taken together, these ten forces capture a field hurtling forward while colliding with real-world institutions—from U.S. privacy law to military doctrine and scientific practice.

How they interact in the coming year will determine whether AI’s next act is defined by breakthrough utility, social retrenchment, or both.