MLB blocks AI-assisted strategy on dugout iPads

Major League Baseball disabled the custom tabs on league-provided dugout iPads as the second half of the 2026 season began. The tablets still provide approved video and league data, but teams can no longer use the custom area to run their own programs during games.

A June 11 memo from MLB executive vice president of baseball operations Morgan Sword said custom apps had expanded beyond the tablets' intended purpose to include recommendations about substitutions, pitch calling and other decisions traditionally made by players and coaches. The Athletic first reported the memo, and AP later said it had obtained the document independently.

This is not a blanket ban on baseball analytics, iPads or artificial intelligence. Teams may still upload static information available before first pitch, subject to MLB review. The restricted capability was the path that could convert live game information into recommendations while a game was underway.

The Athletic's original report and AP's independent memo reporting agree on the core facts: MLB removed access to custom programs, the restriction took effect for the second half, and the league was concerned about software influencing in-game choices. AP also reported that a competition committee review found clubs compliant with existing regulations. The measure should therefore be understood as a preventive governance change, not proof that MLB found teams cheating.

Former pitcher Adam Ottavino separately alleged that the New York Mets had used an expensive AI system to help select pitches. The Mets did not comment. That claim is not necessary to establish the league-wide policy and should remain attributed rather than treated as a finding by MLB.

The league's memo is not publicly available on an official MLB page. That limits the public evidence to the portions independently obtained and reported by the two newsrooms. It also means claims about specific vendors, the precise number of teams using a tool or any undisclosed enforcement action would go beyond what the available evidence supports.

The distinction is useful beyond sports. It separates analysis prepared before a high-stakes operation from model-generated advice delivered at the moment of action. That is often more enforceable than trying to prohibit a broad technology category.

MLB's response is a practical example of controlling AI through the interface where it can influence decisions. Organizations can preserve legitimate analytics while limiting live access to tools, data and actions that create unacceptable risk.

For product and data teams, the comparable controls are narrow runtime permissions, clear separation between precomputed information and live recommendations, durable logs of model inputs and outputs, and a named authority who can disable the integration quickly. A nominal human in the loop is not enough if an opaque system selects the options and delivers them at the exact moment a decision must be made.