Three of the largest quality standards in the world are being rewritten at the same time, the government just forced a rare mandatory recall over a counterfeit safety part, and AI inspection has quietly moved off the pilot bench and onto the production line. None of these is a headline you can file away for later, because each one changes something you are accountable for on Monday. The full conversation is in the episode above. This is the written companion, built for practitioners who want the substance without the fluff.
The standards are converging on purpose
ISO 9001 is expected to publish a new revision around September 2026, and the sector standards are timed to follow it. IATF 16949 is expected to revise after that, tentatively as IATF 16949:2027. AS9100 is being rewritten and rebranded as IA9100 under the IAQG, with publication reportedly expected between late 2026 and mid-2027 and transition windows running toward 2029.
Read the timing, not just the titles. These are not three unrelated projects that happen to overlap. They share a common set of themes: risk management, software and cyber quality, supply-chain resilience, product safety, and, new for ISO 9001, climate considerations. The overlap is deliberate.
If you hold two or three of these certifications, do not run three separate transition efforts. Build one gap analysis that maps the shared themes once, then branch into the sector-specific requirements. Map the common ground first:
- Risk management and risk-based thinking
- Software, cyber, and information-security quality
- Supply-chain resilience and supplier controls
- Product safety and counterfeit-part prevention
Then handle what is unique to each standard on top of that shared base. Organizations that wait for the "final" published text before starting will be squeezed against audit deadlines with no room to maneuver.
A forced recall is a traceability story
On April 29, 2026, NHTSA issued a final decision ordering a recall of aftermarket airbag inflators made by Jilin Province Detiannuo Safety Technology Co. of China. Regulators reported twelve instances where these inflators ruptured and sent metal fragments into occupants, tied to deaths and serious injuries. The parts were reportedly imported and installed as replacement inflators in used vehicles.
The detail that matters for quality engineers is the word "ordering." Regulators almost always rely on voluntary recalls. A forced order is rare, and it tells you the normal controls broke down completely.
This is a traceability and provenance failure as much as a defect. An unapproved part entered the service channel and nobody in the chain caught it before it reached occupants. The controls that stand between you and a failure like this are the ones you already own:
- Approved-vendor lists that are enforced, not just filed
- Certificates of conformance that get verified, not just collected
- Serialization and chain of custody that actually trace back to a qualified source
When a critical safety part moves through the aftermarket, chain of custody is the control that was missing here. Treat incoming inspection and part provenance as safety functions, not paperwork.
AI inspection is here, the hard part is qualifying it
Trade coverage through 2026 describes AI-powered visual inspection crossing from pilot to production at scale across automotive, electronics, pharma, and food, with some systems feeding data back to correct the process in a closed loop. Vendors claim 95 to 99 percent detection accuracy at thousands of parts per hour, and claim human inspection misses 20 to 30 percent of defects with accuracy dropping after about two hours on the line. Treat those figures as vendor claims. They come from vendor and trade content, not from a first-party or peer-reviewed study.
Chasing the accuracy number is the wrong argument anyway. The real quality-engineering question is how you qualify an AI inspector inside a quality system.
- How do you run MSA on a model? A camera-and-model inspector is a measurement system, and it needs the same scrutiny as any gauge.
- What does your control plan say when the model drifts? Drift is not an if, it is a when, and the reaction plan has to exist before it happens.
- Who signs off on retraining? A model that gets retrained is a changed process. That change needs an owner, a validation step, and a record.
"The AI caught it" is not a record an auditor will accept. A documented, validated system behind the model is. Build that system before you put the inspector in a control plan you have to defend.
PPAP is coming to aerospace, and it will break where it already breaks
AS9145 is being revised as IA9145 alongside IA9100. The expected effect is that APQP and PPAP shift from guideline and customer-specific tools to effectively required ones for aerospace compliance, with heavier emphasis on data-driven validation of Key Characteristics, SPC, MSA, control plans, and capability studies.
Here is the collision. Aerospace is a high-mix, low-volume world, and it is inheriting a discipline built for automotive high volume. The capability math assumes volume you often do not have.
The volume problem is real
PPAP capability logic assumes long SPC runs and Ppk on 30-plus parts. Aerospace part families often run in tens, not tens of thousands. A supplier building 12 units a year cannot generate the statistics the form asks for. Forcing a Ppk number out of 8 parts produces a study that is statistically meaningless.
Where PPAP actually fails
PPAP packages rarely fail on manufacturing. They fail at the OEM-supplier handoff, and automotive has lived this for twenty years:
- Ambiguous Key Characteristic flow-down, where the drawing, the control plan, and the MSA disagree about what is critical
- Mismatched revisions between the drawing and the control plan
- MSA run to "pass" rather than to understand the gauge
None of that is a machining problem. It is a communication problem, and it is exactly what aerospace is about to import.
The contrarian take
Copy automotive blindly and IA9145 will produce a wave of compliant paperwork and worse actual quality. The value of APQP was never the binder. It was the cross-functional conversation about risk before anyone cut metal. A low-volume supplier forced to manufacture capability numbers on a handful of parts is producing theater, not control.
The right move is KC-driven, risk-based PPAP. Put deep evidence on the few characteristics that actually matter. Use variables data where volume allows it, and honest attribute data with engineering rationale where it does not. If the aerospace supply base learns automotive's lesson instead of repeating it, IA9145 is a gift. If not, it is a compliance tax.
One habit that pays off now
When you run a Gage R and R, do not grab 10 parts from the middle of today's run. Deliberately select parts that span the full specification tolerance, including a few known near-limit parts.
Gage R and R measures your measurement system, not your product, and the result depends on the sample carrying real part-to-part variation. Cluster 10 parts near nominal and there is almost no part variation, so operator and equipment error looks huge by comparison and a perfectly good gauge fails the study.
Two more things to hold onto:
- Watch the Number of Distinct Categories. NDC should be 5 or greater.
- Separate the two failure modes. Poor repeatability is the gauge itself being inconsistent, which is a hardware or fixture problem. Poor reproducibility is operators using it differently, which is a method or training problem. Each needs a different corrective action, so do not retrain people when the gauge is at fault.
A Gage R and R that passes on ten identical parts didn't prove your gauge is good. It proved you picked easy parts.
The through-line
Every item here rewards the same discipline: reproduce before you conclude, qualify before you trust, and put evidence behind the record instead of paperwork in front of it. The standards are changing, the tools are changing, and the failure modes are stubbornly the same. QualityEngineer.ai builds tooling for exactly this kind of careful, evidence-based quality work. The standards will keep moving. The habits are what hold.
