AI-powered driver help applied sciences have gotten commonplace gear, basically altering how car security is assessed and validated.
The latest recognition of the Mercedes-Benz CLA as Euro NCAP’s Greatest Performer of 2025 underscores this shift, because the car combines conventional passive security options with NVIDIA DRIVE AV software program to attain the best general security rating of the 12 months.
“When Euro NCAP assesses car security, it evaluates each passive and lively programs — attaining an ideal rating requires a state-of-the-art superior driver help system,” mentioned Ola Källenius, CEO of the Mercedes-Benz Group. “This milestone represents the end result of 5 years of collaboration between Mercedes-Benz and NVIDIA to reinforce real-world security and ship tangible worth to clients.”
Euro NCAP (European New Automobile Evaluation Programme) has for almost 30 years served as Europe’s impartial car security authority, backed by European governments, motoring organizations and shopper teams.
Euro NCAP evaluates autos throughout 4 classes that mirror real-world security. For AI-powered driver help, probably the most related are the “Weak Street Person” and “Security Help” classes, which assess applied sciences designed to assist stop crashes — together with automated emergency braking, lane-keeping help and pace help.
Solely autos attaining five-star scores with commonplace gear qualify for “Greatest in Class” recognition, with winners decided by weighted scores throughout all classes. In 2025, Euro NCAP examined a file 49 fashions.
Security Comes First: How DRIVE AV Is Constructed for Belief
Security scores like Euro NCAP are more and more recognizing autos that mix sturdy passive safety with superior lively security efficiency. As AI turns into central to driving, the benchmark for the “most secure” automotive will more and more be outlined not solely by how nicely a car handles a crash, however how successfully it helps stop one.
The Mercedes-Benz CLA is constructed with NVIDIA DRIVE AV, a dual-stack structure that’s designed to assist automakers ship programs that aren’t solely clever, however predictable, verifiable and resilient in the actual world. The structure pairs an AI-driven end-to-end driving system with a parallel classical security stack to supply redundancy throughout AV sensing, planning and execution.
The CLA can also be constructed on the NVIDIA DRIVE Hyperion structure, which contains sensor variety and {hardware} redundancy into the car’s general design.
On the coronary heart of this method is NVIDIA Halos — a complete security system spanning {hardware}, software program, instruments, growth processes and certification help. Halos delivers a structured security basis for growing automated driving and different AI capabilities whereas staying anchored to strong guardrails, redundancy and fault tolerance.
Third-party certification and assessments are additionally necessary to construct belief:
- TÜV SÜD granted the ISO 21434 Cybersecurity Course of certification to NVIDIA for its automotive system-on-a-chip, platform and software program engineering processes. Moreover, NVIDIA DriveOS 6.0 conforms to ISO 26262 Automotive Security Integrity Stage (ASIL) D requirements.
- TÜV Rheinland carried out an impartial United Nations Financial Fee for Europe (UNECE) security evaluation of NVIDIA DRIVE AV associated to security necessities for complicated digital programs, which NVIDIA efficiently accomplished.
NVIDIA just lately launched its Alpamayo household of open AI fashions, simulation instruments and datasets — which allows AVs to navigate even uncommon, “long-tail” occasions they haven’t been educated on by breaking the situation down into smaller steps, reasoning via a number of potential actions and finally deciding on the most secure one. Utilizing these fashions with the parallel classical security stack within the NVIDIA DRIVE AV dual-stack structure offers an extra layer of safety to maintain autos working inside secure boundaries.
Coaching Security By way of Knowledge and Simulation
Fashionable AI-driven security programs study from exponentially extra driving situations than any human might expertise in a lifetime. NVIDIA’s cloud-to-car growth method transforms real-world knowledge into billions of simulated miles utilizing NVIDIA DGX programs for neural community coaching, the NVIDIA Omniverse and Cosmos platforms for simulation, and NVIDIA DRIVE AGX for in-vehicle computing.
This system addresses a crucial problem in security validation: coaching AI to navigate uncommon however high-risk edge circumstances which are too harmful — or too rare — to check reliably in the actual world. By producing artificial situations that characterize these uncommon conditions, AI programs can study applicable responses throughout growth with out placing folks in danger.
The CLA’s recognition is greater than a single mannequin incomes a high score — it displays a broader shift in what security means within the trendy car, the place trusted crash safety is paired with AI-enabled driver help designed to assist keep away from accidents within the first place.