Sarah PetersonThe San Diego Padres beat the Washington Nationals 3-2 on September 7, 2026, at Petco Park — but the most talked-about moment of the series was not Nick Pivetta's five shutout innings or a late-inning home run. It was a batter tapping his helmet twice, triggering a challenge to the umpire's ball-and-strike call via baseball's new Automated Ball-Strike (ABS) Challenge System. The challenge was upheld. The pitch that had been called a strike was ruled a ball by the algorithm. The at-bat continued. The game, ultimately, shifted.
Across the 2026 MLB season, 53% of challenged calls to the ABS system have been overturned — meaning the automated system disagreed with the human umpire in more than half the cases where a player or catcher decided to push back. For a league that has used human judgment on balls and strikes for 150 years, this is extraordinary. For businesses deploying their own AI decision systems, it is a warning wrapped in a baseball story.
What Is the MLB ABS Challenge System — and How Does It Work?
The ABS Challenge System debuted on Opening Day of the 2026 MLB season. Across all 30 ballparks, including Petco Park where the Padres hosted the Nationals this past weekend, twelve Hawk-Eye cameras are synchronized to track every pitch to sub-millimeter accuracy. The network runs on a private 5G infrastructure from T-Mobile, designed to deliver real-time processing with minimal latency.
When a batter, pitcher, or catcher believes the home plate umpire has misclassified a pitch as a ball or a strike, they can signal a challenge by tapping their helmet or cap. The call goes to the ABS system, which processes the pitch trajectory against a digital model of the strike zone and returns a verdict in seconds. The human umpire announces the result.
Each team gets exactly two challenges per nine-inning game. Additional challenges are granted in extra innings. Coaches, managers, and other players on the field cannot trigger a challenge — only the batter, pitcher, or catcher at the time of the pitch.
The system's technical performance during 2025 spring training testing was formidable. Of 88,534 pitches tracked, just four went undetected. Of 1,214 challenges tested, five failed to display a result. That translates to a system reliability rate above 99.99%.
On paper, this is extraordinary engineering. In practice, the debate begins exactly where the engineering ends.
Why Do Humans Keep Winning the Argument — and Why Does That Matter?
The 53% overturn rate is the statistic that most people find surprising, and it cuts in both directions. Baseball purists argue it proves the ABS system is too aggressive — overriding experienced umpires on borderline pitches that require contextual judgment. Technology advocates counter that it reveals just how frequently human umpires have been getting high-leverage calls wrong for decades.
Both interpretations miss the more important question: what does the ABS system actually measure, and what was it designed to measure?
The system uses a two-dimensional representation of the strike zone, calculated from the batter's height and stance at the start of the at-bat. Human umpires have historically applied three-dimensional judgment — accounting for pitch trajectory, late break, and catcher framing. When a pitch crosses the front edge of the plate at one height and the back edge at another, the ABS system makes a call based on geometry. The umpire makes a call based on a lifetime of watching ball flight.
The margin of error in the system — approximately one-sixth of an inch — is smaller than a baseball lace. But it is not zero. And the pitches that generate the most challenges are precisely the pitches where that sixth-of-an-inch matters: full-count at-bats, two-out situations, late innings in one-run games. The September 7 Nationals-Padres game drew two challenges in situations that directly affected batter count progressions, changing the outcome of at-bats that contributed to the Padres' 3-2 margin.
Only 1.4% of all 2026 pitches have been challenged. But those challenges cluster in high-stakes moments, making the edge cases the most consequential.
What the ABS Debate Reveals About AI Decisions in Your Business
Major League Baseball has resources that virtually no business in America can match: a decade of development, twelve cameras per venue, private 5G infrastructure, a Hawk-Eye system refined from professional tennis, and a reliability rate of 99.99%. And still, the algorithm generates controversy because it measures something different from what experienced humans were measuring — and no one fully anticipated where those differences would matter most.
This is the central lesson for any organization deploying an AI decision system.
According to the NIST AI Risk Management Framework, responsible AI deployment requires organizations to document not just whether a system is accurate overall, but how it performs on edge cases, who bears accountability when it errs, and what problem it was actually designed to solve. The ABS system was built for a controlled, standardized environment — same ball, same field dimensions, same challenge protocol every game. Most business AI deployments operate in far messier conditions, and most organizations discover the edge cases only after deployment, not before.
An IT consultant specializing in AI systems can help your organization map those edge cases before they become costly. That is not a sales pitch — it is the lesson that MLB, with all its resources, is still working through in real time.
The Real Scenario: What Happens When Your AI System Gets the Call Wrong
Consider a mid-size manufacturing company in Pennsylvania that installs a machine-vision AI system in early 2026 to flag defective components on its assembly line. The vendor's documentation promises 97% accuracy — meaningfully higher than the 91% accuracy of the best human inspector on the team. The company invests $140,000 in installation and integration.
Six months into deployment, the quality team notices that the system flags approximately 9% of components as defective. When managers pull a sample of flagged items for manual inspection, they discover that 4.5 percentage points of that flagged group — half the flagged volume — consists of components that are perfectly functional. The system is systematically over-calling defects on parts manufactured during afternoon shifts, when ambient light in the facility shifts as sunlight enters through skylights.
The AI was trained on images captured under standardized artificial lighting. No one told it that real-world lighting conditions in this building change by season and time of day.
If the system is discarding or quarantining 4.5% of total production as false positives, and the plant produces 28,000 components per month at $14 per unit, that is $17,640 in falsely rejected inventory every month — or roughly $211,680 per year. The 97% accuracy claim was accurate in the training environment. It was not accurate in the deployment environment.
An IT consultant with AI systems experience would audit the training data composition, identify the lighting variable as a confounding factor, and either recommend recalibration with new training data representing the full range of production conditions, or a conditional override protocol during high-ambient-light periods. Without that audit, the company discovers the problem only through accumulated financial loss — the equivalent of a team spending the entire season losing challenges it should have won.
The ABS system in baseball is essentially a built-in challenge mechanism — players can surface disagreements with the algorithm two times per game. Your business AI system almost certainly does not have a comparable escalation pathway built in. You need an IT professional to design one.
Who Should Be in the Room When an Algorithm Makes the Decision?
As the Nationals-Padres analytics matchup this season illustrates, the teams that use data systems most effectively are not necessarily the ones with the most technology. They are the ones that understand which decisions belong to the algorithm and which belong to the human — and they structure their workflows accordingly.
The ABS system does not replace the umpire. It checks the umpire. That design choice is deliberate: MLB's architects acknowledged that the system performs best within a clearly defined measurement domain, and they kept a human in the loop as both announcer and contextual backstop. Most enterprise AI deployments invert this relationship — the algorithm produces an output, and the human receives it without access to the reasoning.
An IT expert can help your organization define the boundaries of what your AI system should decide autonomously and what should require human review. In regulated industries — healthcare, financial services, logistics, insurance — this distinction is increasingly formalized by state and federal algorithmic accountability frameworks that establish disclosure and audit requirements for automated decision systems.
Three questions to answer before any AI system goes live:
Can you reconstruct any individual decision? If you cannot explain why the system produced a specific output for a specific input, you cannot audit it, correct it, or defend it. Logging and traceability are not optional.
Where does the system perform worst? A 97% accurate system fails 3% of the time — and that 3% is rarely random. Identify the clustering before deployment, not after.
What is the escalation path? The ABS system answers challenges in seconds, publicly, with the umpire announcing the result. Your AI system needs an equivalent accountability chain: who reviews a contested decision, how quickly, and what documentation exists.
What to Do Now
The Nationals-Padres series made robot umpires front-page news. But the practical questions the ABS debate surfaces — about accuracy at the edges, accountability when errors occur, and the limits of algorithmic judgment — are questions every business deploying AI should be asking in 2026.
ExpertZoom connects organizations with IT specialists who audit automated decision systems, review AI accuracy across real-world deployment conditions, and design accountability frameworks that protect you when the algorithm gets the call wrong. Because unlike in baseball, you do not get two challenge tokens. You get the outcome, and you need to be ready for it.
