There’s a delicate art to teaching robots, even when you’re preparing them for predictable environments like factories, where they’ll repeat the same tasks a little differently depending on the obstacles they face. Whether a human is suddenly in their way or there’s new clutter, the machine must closely mimic its operator’s actions by staying on a trajectory (or motion path).

When security researchers want to understand what a modern processor is really doing with the kind of detail that determines whether attacks like Spectre and Meltdown are possible, they usually run their experiments on top of an operating system that was never built for the job. They open up macOS or Linux, patch the kernel by hand, and hope the modifications hold. The approach is unstable, hard to reproduce, and on Apple's platforms, slated for deprecation.

The multiplication of matrices and of higher-dimensional arrays called tensors lies at the heart of modern computing. Matrix or tensor multiplication is, in fact, the most common operation carried out in artificial intelligence applications, as well as in scientific simulations and computer graphics. 

For MIT Professor Armando Solar-Lezama, one of the most common misunderstandings about AI is the notion that it can be dropped into existing human roles like a plug-and-play replacement.

In 2026, the hype for artificial intelligence (AI) agents is louder than ever before. These semi-autonomous programs can “think” and execute well-defined tasks in areas like customer service and software development, typically using language models (LMs). But fields like medical diagnosis and scientific discovery require them to inquire about a vast range of solutions in uncertain environments, which LMs struggle with.

A new method developed by MIT researchers can accelerate a privacy-preserving artificial intelligence training method by about 81 percent. This advance could enable a wider array of resource-constrained edge devices, like sensors and smartwatches, to deploy more accurate AI models while keeping user data secure.

Confidence is persuasive. In AI systems, it is often misleading.

David Clark, Senior Research Scientist at MIT CSAIL, helped design the system that connects nearly every computer on earth. As Chief Protocol Architect of the Internet from 1981 to 1989, he was there for the beginnings of the Internet. Calling the wave of AI technology an “echo” of what happened in the 80’s, Dr. Clark is cautioning, “ maybe we need to slow things down and think a bit.”

What happens when the team behind PlayStation meets the researchers pushing the boundaries of AI? You get The Nexus of Games and AI, a 12-part MIT Independent Activities Period (IAP) course, now available to stream.

AI models are proliferating fast. There’s Claude, ChatGPT, Gemini, Copilot, DeepSeek, Grok, Mistral, Llama, and many more emerging every day. But which ones to work with? And why? We asked MIT CSAIL faculty and students which AI tools they’re reaching for right now. The responses showed a variety of preferences, a clear winner in one area, and a word of caution about what goes into any public model’s memory.