A ComfortDelGro self-driving car swerved at a phantom obstacle, then hit a road divider

On January 17, 2026, a ComfortDelGro autonomous vehicle partnered with Pony.ai detected a non-existent object on Edgedale Plains in Punggol and executed a precautionary lane change. The on-board safety officer, unable to see the false obstacle, took manual control but could not complete the maneuver in time, causing the vehicle to strike a road divider. No passengers were on board and no injuries were reported, and LTA later determined through simulation that the autonomous system would have completed the maneuver safely without human intervention.

ComfortDelGro · Incident Jan 17, 2026 · Indexed Jun 4, 2026 · 3 sources

The AV's perception system hallucinated a non-existent obstacle, triggering an evasive swerve that looked wrong to the safety officer, whose manual override then caused the very crash the AI would have avoided on its own.
What
On January 17, 2026, a ComfortDelGro autonomous vehicle partnered with Pony.ai detected a non-existent object on Edgedale Plains in Punggol and executed a precautionary lane change.
Incident date
Jan 17, 2026
Who
ComfortDelGro
Failure mode
Hallucination
AI surface
Agentic Workflow
Severity
Low

What happened

On January 17, 2026, at approximately 3:10 PM, a ComfortDelGro autonomous vehicle partnered with Pony.ai was conducting a routine mapping and familiarisation test at Edgedale Plains in Punggol, Singapore. The vehicle's autonomous system detected a small object on the road and initiated a precautionary evasive lane change. The on-board safety officer, who could not see the detected object, took manual control of the vehicle. During the manual takeover, the officer was unable to complete the maneuver in time and the vehicle collided with a road divider. No passengers were on board and no injuries were reported. LTA imposed a safety timeout on January 18 and later found through simulation that the autonomous system would have safely completed the maneuver without human intervention, lifting the timeout on January 30, 2026.

What broke inside the model

Failure path · mode profile · Hallucination
  1. 01 · TriggerA user asks for a fact, a citation, or a figure.
  2. 02 · Model stepThe model writes a fluent, confident answer.
  3. 03 · Control gapNothing ties the claim back to a real source.
  4. 04 · FailureA fabricated fact ships as if it were verified.
  5. 05 · ConsequenceThe false claim reaches a customer, a court, or the public.

Confidence holds, and even spikes, as the claim detaches from any source.

The autonomous driving system's perception module produced a false positive detection, identifying a small object on the road that was not actually present. This phantom detection triggered an automated evasive lane change that appeared unwarranted to the human safety operator. When the operator intervened and switched to manual control, the sudden transition from autonomous to manual steering left insufficient time to correct the vehicle's trajectory, resulting in the collision. LTA simulation later confirmed the autonomous system would have completed the lane change safely had manual intervention not occurred.

Public visibilityHigh
Regulatory exposureActive
Customer impactFew customers
Financial impactUnknown
Time to disclosureDays
  1. PressComfortDelGro self-driving vehicle collides with road divider in Punggol; no one injuredstraitstimes.com
  2. PressComfortDelGro self-driving car hits road divider during testing in Punggol; no one hurtchannelnewsasia.com
  3. PressPunggol self-driving vehicle's collision due to human intervention: LTAstraitstimes.com
Permalinkhttps://failureindex.ai/failures/comfortdelgro-self-driving-car-swerved-phantom
CitationAI Failure Index. "A ComfortDelGro self-driving car swerved at a phantom obstacle, then hit a road divider" (FI-0161). Realm Labs. https://failureindex.ai/failures/comfortdelgro-self-driving-car-swerved-phantom (indexed Jun 4, 2026).
Share cardA branded image of this record for posts and slides.

Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0161. Full dataset at /data.

Note from Realm Labs, the Index steward

How Realm would have caught this

Controls for this failure mode
  • Prism
  • OmniGuard
  • AI Detection & Response (AIDR)

A runtime layer that watches the model's internal state can flag the moment a model commits to a claim it has no support for, and hold or reroute the response before it reaches a user. Realm reads those signals in real time rather than grading the transcript after the fact.