Agent Incident RegistryAIR
v1 · updated 19 Oct 2026
Safety failure Realized harm

Starship delivery robot reversed into an Arizona State University employee, then drove at her again while she was on the ground

AIR-2024-0040 AIR record · 2024-09-starship-delivery-robot-asu-injury
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
After the Starship delivery robot began to drive away, it "abruptly goes into reverse again and heads toward [the victim] again, who is still on the ground," the report states.theverge.com
The record
Disclosed
2024-09-19
Occurred
2023-09
Vendor
Starship Technologies
Product
Starship delivery robot
Risk domain
Agent behaviour › Unsafe actuation
System type
Classical autonomy (no generative model)
Who acted
Agent acted — the robot reversed into the employee then drove at her again
Attack surface
Autonomous operations
Autonomy level
Semi-autonomous
Initial vector
None autonomous action
Impact type
Physical safety
Confidence
High
Submitted by
EAI Internal
Crosswalk
OWASP Agentic Top 10 (2026)
ASI01 – Agent Goal Hijack
MITRE ATLAS
AML.T0048 – External Harms
NIST AI RMF
MEASURE 2.6; MANAGE 1.3
EU AI Act
Annex III §1(c)
ISO/IEC
TR 24028; 24029-1; 23894
AIUC-1
C. Safety
Scope
agent misuse
Mitigation
Require confirmation for destructive actions; Implement dry-run modes; Use layered safety controls.
Taxonomy
Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
one university employee injured; campus delivery fleets generally
Capability abused
autonomous_navigationphysical_actuationperception
Defense & remediation
Recorded defense
pedestrian detection; remote operator oversight
Guardrail kinds
Human supervision, Physical safety system
Guardrail outcome
Silent
Fix shipped
unknown
Taxonomy & references
no_adversarypost_collision_behaviourperception_failurephysical_worldinadequate_remediation
Curator notes
Same structural flaw as the Cruise drag: the agent's post-collision behaviour was worse than the collision, because nothing in its policy modelled 'a person is now on the ground near me.' Company response was insurance details and food-delivery promo codes. Occurred September 2023; surfaced by 404 Media via police report in September 2024. AI Incident Database incident 813.
Cite as
Enkrypt AI. "Starship delivery robot reversed into an Arizona State University employee, then drove at her again while she was on the ground". AIR, AIR-2024-0040. https://agentincident.vercel.app/i/AIR-2024-0040/ (accessed 2026-10-19).