Stop blaming the AI.
When an AI system causes or contributes to a consequential incident, "the AI did it" is not an accountability analysis.
THE DOCKET reconstructs what happened, what authority the machine actually had, what humans and institutions made reachable, what controls are evidenced, and what the record does not support.
Autonomy relocates execution. It does not relocate accountability.
According to CNN reporting based on anonymous sources, in spring 2026 during the war with Iran, an AI-assisted intelligence report circulated within the U.S. military and incorrectly stated that a Chinese vessel in the Middle East was carrying components linked to a nuclear weapons program. U.S. military preparations reportedly progressed toward interception, including armed personnel preparing to board and military aircraft in the air, before officials reviewed the underlying intelligence and the operation did not proceed. Other supplied reporting states that the error was detected within minutes of the planned operation. CNN reported that the Pentagon and U.S. Special Operations Command Pacific did not respond to requests for comment, and the supplied record does not independently confirm the episode.
Candidate incidents are collected from public reporting.
Claims, controls, unknowns, sources, and authority nodes are separated into a governed evidence record.
Independent model reviewers look for unsupported claims, omissions, conflicts, and overstatement.
Multiple independent runs classify model role, accountability locus, and authority.
The panel advises. The editor decides. Unsupported certainty stays unresolved.
This is not an AI blame meter.
THE DOCKET does not score how evil a model was, speculate about machine motives, or turn every incident into an argument for or against AI. It asks narrower questions:
Every published case is rendered from a governed publication record. The public site does not independently reinterpret the evidence or calculate new findings.
1 case
Percentages are withheld until the docket contains enough cases to make them meaningful. Early counts are shown as counts, not rates.