Biometric security systems at three Tier 4 corponation campuses began flagging anomalies in their executive authentication protocols in early 2225. The flags were not intrusion alerts. They were something the systems had no classification for: identity matches that the AI simultaneously validated and rejected. DNA profiles matched. Retinal patterns matched. Voice prints matched. Gait analysis matched. Every measurable biometric parameter confirmed that the individuals passing through security were who they claimed to be. And the AI, running behavioral prediction models trained on years of individual data, flagged them as copies. Not impostors. Not synthetics. Copies — the same person, arriving at the door, but somehow not the original.

The three affected executives — one at Palladian, one at Nakamura-Holt, one at a Tier 4 logistics conglomerate called Venn-Strata — were flagged independently within the same two-week period. Security teams at each campus ran full diagnostic cycles. Hardware was replaced. Software was reinstalled from verified backups. The AI continued to flag the executives. Human security personnel were brought in to evaluate. They saw nothing wrong. The executives looked right, sounded right, knew everything they should know. When shown the AI's confidence scores — 99.97% biometric match, 0.3% behavioral match — the human evaluators overrode the system. The executives were cleared. The AI was noted as malfunctioning. It was not malfunctioning.

The originals have not been found. This is not a statement the documentation project makes lightly. Investigation into the three flagged executives reveals no moment of substitution, no gap in surveillance coverage, no window during which a replacement could have occurred. The executives appear to have always been who they are now. Family members notice nothing. Long-term colleagues notice nothing. The discontinuity exists only in the AI's behavioral models, which insist that the executives' decision-making patterns, micro-expressions, and stress responses shifted simultaneously on a date in January 2225 that none of the three executives can account for. Their calendars show normal workdays. Their badge logs show normal movement. But the AI sees a seam — a point where one pattern of being ended and another, nearly identical but not quite, began.

Performance metrics for all three executives improved markedly after the flagged date. Palladian's executive restructured a failing supply chain division in six weeks — a task that had stymied her predecessor personality for two years. Nakamura-Holt's executive resolved a labor dispute by making concessions that the pre-January version would never have considered. Venn-Strata's executive launched a charitable initiative that industry analysts described as "uncharacteristically human." In each case, the improvement was noted, celebrated, and not questioned. One executive assistant — Palladian, name withheld — described the change in her boss with a precision that the documentation project found unsettling: "She's better now. Kinder. More decisive. Sleeps less. I should be more worried than I am. I know I should be more worried than I am. I can't seem to get there."

The GLMZ Anomaly Documentation Project does not have a classification for this event. The executives are real. Their identities are verified. They perform their roles effectively. They appear to be the same people. The AI says they are not. The question of which observer to trust — the human eye or the machine pattern — is not philosophical in this case. It is operational. Three of the most powerful individuals in the GLMZ were replaced, or transformed, or optimized, by a process that left no trace except a statistical anomaly in a behavioral model. The copies, if they are copies, are better than the originals. And no one with the authority to investigate seems able to want to.