Turing says autonomous-driving model learned to flash hazard lights to say “thank you”

jidounten-lab.com reported on Oct. 3, 2026, that Turing’s autonomous-driving model had learned to flash its hazard lights to say “thank you.” According to the article, founder and CEO Kazunari Yamamoto posted on X on Sept. 26, 2026, and Director and COO Daisuke Tanaka reposted a similar message that same day.

The report appears to show Turing presenting autonomous driving not simply as route following, but as behavior that includes driving culture and consideration for other people. Whether behavior around the vehicle can be confirmed in demonstration videos, in addition to driving functions, is directly relevant to user acceptance and safety assessments.

The focus of the report is the claim that Turing’s autonomous-driving system demonstrated not only driving tasks such as lane keeping and merging, but also a peripheral driving behavior: turning on the hazard lights after merging. However, this is not a case in which performance metrics were provided through public data or an official announcement; the report is based on posts on X and a video, so readers should consult the original sources for details. Related coverage from Physical AI News includes Waymo launches accounts for teens in Nashville, shares location information with parents (Sept. 28, 2026), which covered operational design involving not only driving itself but also user management and location-sharing features for parents. The difference here is the claim that the driving system displayed a gesture resembling human behavior. Waymo and Zoox report injuries in public-road tests (Sept. 27, 2026) focused on safety in public-road testing; the issue here is less safety itself than how far a model can incorporate customs that are not formal rules. Along with Aseon Labs reportedly plans to automate vehicle maintenance (Oct. 2, 2026), the coverage points to a shift in the basis of competition in autonomous driving, from driving algorithms alone to operations, maintenance and behavior toward other people. In terms of its technical approach, Turing’s publicly disclosed end-to-end (E2E) autonomous-driving model is designed to learn from input to output as a unified system, so it would not be surprising if behavior such as flashing hazard lights emerged as part of the driving outcome. On the other hand, flashing hazard lights to say “thank you” is not legally required and is strongly tied to Japanese driving culture. Assessments may therefore differ depending on whether the model learned to comply with traffic laws or to imitate a peripheral driving custom. Public information does not establish whether this was an intentional feature addition or an incidental behavior arising from the training data. Three points remain to be confirmed. First, whether similar behavior can be reproduced under multiple conditions, rather than in a limited video. Second, in what driving situations the hazard lights come on, and whether the system is designed to avoid false activations or misunderstandings by other drivers. Third, how far Turing will officially characterize this as a feature resulting from training its E2E model.

In Japan, flashing hazard lights to say “thank you” is a custom that is generally understood, making it relevant to consider how much a self-driving system trained on domestic driving contexts, such as Turing’s, will incorporate human customs. Whether this can become a product feature is a separate question and requires evaluation from both regulatory and operational-design perspectives.