AI Wrote Your TARA. Who Signs It? (10 July 2026)

By DNSystems LLC (dnsystemsllc.com/ptrg)

ISO/SAE 21434 · UN R155 · TARA vs HARA · Penetration Testing as a Service

👉 Sample report bundles: https://ptrg.dnsystemsllc.com/sample-reports

https://youtu.be/8vlIZJqFz0c

Every automotive-security vendor now ships an AI that drafts your Threat Analysis and Risk Assessment. FEV's TARA Copilot takes a Word item definition and emits an Excel workbook through a five-module LLM chain. PlaxidityX's AutoDesigner generates one from your architecture model. PTRG's DeepCore co-pilot drafts the full ISO/SAE 21434 Clause-15 analysis from a block diagram in a single pass. The draft, in other words, is no longer where the value lives. So where does it?

What a TARA actually is — and how it differs from HARA

A TARA is the ISO/SAE 21434 Clause-15 workflow for road-vehicle cybersecurity: define the item and its assets, derive damage scenarios and impact ratings across safety, financial, operational and privacy dimensions, enumerate threat scenarios and attack paths, rate attack feasibility, compute a risk value, and then decide a risk treatment — mitigate, share, retain, or avoid.

It is often confused with HARA, the Hazard Analysis and Risk Assessment defined by ISO 26262 for functional safety. HARA asks "what happens if this system fails or malfunctions?" and produces an ASIL rating. TARA asks "what happens if an attacker makes this system misbehave on purpose?" and produces a cybersecurity risk. Modern programs run both, because a malicious CAN injection and a random bit-flip can drive the same unsafe vehicle behaviour — but only one of them is deterred by a signature, a MAC, or a secure boot chain. Confusing the two is one of the most common ways a compliance program leaves an attack path unassessed.

The tell: automation stops at determination, not treatment

Read the fine print on the AI-TARA tools and a pattern emerges. FEV's own material is candid: risk determination is automated; risk treatment is not. You receive a spreadsheet, and the decision about what to do with each threat stays manual. That boundary is not an accident — it is the point at which an assessor stops trusting a machine and starts wanting a human who will put their name on the outcome.

This matters because of what sits on the other side of the TARA. UN Regulation No. 155 makes a certified Cybersecurity Management System a gate to vehicle type approval across every contracting party to the WP.29 1958 Agreement — no compliant, maintained TARA, no sale. The CSMS certificate is valid for three years, and renewal requires evidence that the process was applied to the current vehicle configuration, not the one you shipped at launch. A spreadsheet of undecided risks does not clear that bar.

Where PTRG picks up: treatment, proof, and a signature

PTRG uses the same AI to draft assets, damage scenarios, threats, attack paths and feasibility. Then it does the three things the draft cannot:

• Treatment, decided. A countermeasure is mapped to every threat, with per-finding NIST 800-53 / 800-171 (and gated 800-193 firmware-resiliency) plus UN R155 mitigation references — not just a risk value left hanging.

• Exploitation, proven. Where it matters, DeepStrike runs live, reproducible exploitation on the bench and attaches the proof-of-concept evidence behind the risk score. A public vulnerability lowers attack effort, so the risk is re-scored on evidence, not assertion.

• Deliverable, signed. The output is a countersigned, AES-256-sealed bundle — PDF, DOCX, slides and CSV — re-rendered in under ninety seconds whenever the design, the threat landscape, or a component's vulnerabilities change. Re-assessment is not a new five-figure engagement.

👉 Sample report bundles: https://ptrg.dnsystemsllc.com/sample-reports

This is Penetration Testing as a Service (PTaaS) applied to automotive: the assessment, the exploitation evidence, and the signed compliance artifact arrive as one continuous product rather than three disconnected engagements.

The same pattern shows up beyond the car

The design-time-analysis-versus-signed-evidence gap is not unique to automotive. In medical devices, FDA §524B has required a machine-readable SBOM and a vulnerability-handling plan in premarket submissions since 2023. The EU Cyber Resilience Act extends lifecycle vulnerability handling to every product with digital elements, with vulnerability reporting obligations from September 2026. NIS2 and DORA push the same expectation into operators and financial entities. In each regime the regulator is asking the same question the automotive assessor asks: not "did a tool flag it?" but "who analysed it, who decided what to do, and can you prove it?"

The honest comparison

We publish the whole field side by side in one honest grid — Block Harbor, ThreatZ, PlaxidityX, FEV and PTRG — and we mark where each one leads. Everyone can draft. PTRG covers your entire attack surface (web, API, network, cloud and automotive), decides the treatment, proves the real risks with live exploitation, and hands you a signed deliverable at flat, published pricing — on an open bench stack with no proprietary license tax.

AI writing the draft is table stakes. Signing the deliverable — and proving it — is the job.

See the module-by-module comparison at ptrg.dnsystemsllc.com/vs/fev, the whole field at ptrg.dnsystemsllc.com/vs, and score your own TARA maturity free in sixty seconds at ptrg.dnsystemsllc.com/compliance/readiness.

— DNSystems LLC

PTRG. For Pentesters, By Pentesters.