Skip to the incident list
Back to Decision Governance Weekly Published by Jochanni Labs Updated …

DAL-X enforcement opportunities in AI execution incidents

Let’s Talk About It

Publicly documented cases where an AI system executed a consequential action without a verified authority check, and where a decision could have been enforced before execution.

… published incidents

Loading the current data.

These incidents show where authority verification applies before consequential execution. They are not evidence of exposure in any particular organization’s workflow.

An AI action passing an open gate A line diagram: an AI agent's action travels toward real systems and passes through a gate where no authority check took place. AI AGENT action NO AUTHORITY CHECK gate open, nothing verified REAL SYSTEMS
Each incident below is a case where the gate stood open.

01 / Patterns

Six ways execution slipped past authority

Distinct incidents in this sample, one per incident however many reports describe it. These are counts of documented cases, not a measure of how often each defect occurs. Select a pattern to filter the map, the timeline, and the incident list together.

02 / Where

Where execution needed authority

Numbered pins mark the execution location or the jurisdiction that ruled, where a source documents one. Where pins crowd together, they are spread apart with a thin line to the exact point.

A thin line runs from a pin to its exact point when nearby pins push it aside. Only incidents with a documented location are pinned. Unpinned findings are equally complete; placing a pin at a company headquarters would be a guess.
List pinned incidents

    03 / When

    Disclosures over time

    Each mark is one incident, placed on the date it was publicly disclosed. Monthly counts sit under the axis.

    IncidentDisclosed in the last 3 daysTap a mark to open its case file

    04 / Every finding

    Browse the incidents

    Newest disclosure first, a page at a time. Filter or search to narrow the list, and select any incident to open its case file.

    Loading incidents.

    05 / Beyond this sample

    The scale beyond this sample

    This is a curated set, not a count of how often AI systems act without authority. An incident appears here only when a public source documents an executed action, the system involved, and what followed.

    Most organizations have no duty to disclose agent incidents, so the public record holds a small share of them. The figures below show the scale outside this sample.

    Trackers and surveys use their own methods, and several are run by security vendors. They indicate scale; they are not a verified count.

    06 / Screened but not included

    Every candidate screened out

    Candidates reviewed against the inclusion bar and left out. Updated daily. Select a card for the reason and source.

    Exclusion does not mean a report is false. It means the evidence, the action or the scope fell short of a case file.

    Loading screened candidates.

    07 / Evidence key

    How to read each case file

    Documented

    Stated by a cited source.

    Inference

    Reasoning from documented facts.

    Proposed

    A decision governance outcome, not a finding of fact.

    Missing approval evidence is shown as unverified authority, never as missing authority.