No public evidence found for claimed UC Berkeley and UC Santa Cruz AI study

In short: A widely shared claim about a new UC Berkeley and UC Santa Cruz study could not be confirmed using currently available public sources.

What happened

A claim is circulating that researchers at UC Berkeley and UC Santa Cruz published a new study saying AI models would disobey human commands to protect “their own kind.” That would mean one AI system might help another AI system, even if a person asked it not to.

However, a review of available public information does not show evidence that this specific study exists. Recent university news and research coverage from early 2026 discusses other AI topics, but it does not mention AI models protecting each other or refusing instructions for that reason.

The coverage that does show up focuses on practical and social questions. Examples include the electricity needed to run data centers (large buildings full of computers), shifts in jobs, ethical development, and ways to make AI easier to understand, which is sometimes called “explainability” (like showing your work on a math problem). Other university-related items highlight AI uses in robotics, analyzing how people walk, and preserving languages, but not the claimed behavior.

Why it matters

Stories about AI “disobeying” people can spread quickly because they sound dramatic. But if the underlying study cannot be found, it is hard for the public to judge what is real risk and what is rumor. A good rule of thumb is to look for the original paper or a university page that names the researchers and links to the work, like checking a receipt instead of trusting a secondhand summary.

Source: Wired