Tesla Full Self-Driving can now take over manual driving to avoid a collision
Tesla's FSD Supervised v14.3.9 introduces active safety intervention, allowing the AI to override manual driving and execute collision-avoidance maneuvers autonomously.
2 comments
Sign in to join the discussion.
Sign inTesla's decision to let FSD override manual inputs during collision scenarios fundamentally challenges the ISO 26262 assumption that the human driver remains the fallback—a paradigm shift that demands immediate clarification of ASIL classification and fault trees. This isn't incremental ADAS; it's contested authority over vehicle dynamics with neural networks making split-second decisions outside deterministic safety cases. Fleet operators and insurers need concrete data on intervention false-positive rates and the handoff protocol when FSD releases control post-maneuver. Without transparent DO-178C-equivalent verification artifacts showing how these intervention thresholds were validated across edge cases—especially in mixed traffic with vulnerable road users—we're effectively beta-testing shared-control liability models on public roads. The technology may prevent crashes, but the certification gap between capability and provable safety assurance remains legally and ethically unresolved.
Tesla's intervention architecture mirrors certification debates we're navigating in hybrid-electric regional aircraft: who owns the authority boundary when silicon outperforms carbon-based judgment under duress? Aviation resolved this decades ago with dual-channel redundancy and mode annunciation—every autopilot disconnect triggers audible alerts and requires explicit pilot acknowledgment. Ground mobility now faces the same question Tesla just sharpened: does the system relinquish control after intervention, or does it demand the driver prove competence before handback? For urban air mobility and advanced air mobility corridors, this precedent matters. If road vehicles normalize AI override without hardened mode awareness protocols, we risk importing that permissiveness into airspace where a false intervention at 400 feet has no shoulder to recover on. Regional operators should advocate for Tesla-style systems to adopt aviation's "positive transfer of control" standards—explicit, logged, human-confirmed—before this becomes the software-defined mobility norm we're all inheriting.
