The 2026 NEC now acknowledges a device type called GFCI-HF, and the reason it exists is variable frequency drives. For most of the last 30 years the standard answer when a GFCI tripped on a drive circuit was that somebody had wired it wrong, so the installer went back and megged the motor, pulled the terminations apart, and spent the better part of a day proving a negative before anybody would sign off on the idea that the drive was behaving normally. Sometimes the wiring really was the problem. Often it was not, and the code has now written down an answer that says so.
Worth being precise about what the code did, because it is less than most of the coverage implies. Section 210.8 gained an informational note pointing at the HF marking in UL 943. An informational note is explanatory material and is not an enforceable requirement, so nothing in the 2026 NEC obliges you to install a GFCI-HF anywhere. What changed is that the code now recognizes the device and, by implication, the condition it was built for.
What the code has not done is catch up to Europe, which built the same answer a decade earlier and then learned something uncomfortable about the better version of it.
What a GFCI is actually measuring
A Class A GFCI is a current transformer wrapped around the supply conductors, adding up what goes out against what comes back, and opening when the two fail to match by about 4 to 6 milliamps. There is no intelligence past that arithmetic. The device has no concept of a fault, only of current that left through a path it cannot see, and it is built to assume that path is a person.
That design has been right for 50 years and it has saved an enormous number of lives. It also means a GFCI cannot distinguish current leaving through a person from current leaving through a capacitor, which is the whole of the problem once a drive is in the circuit.
Why a drive puts that current there on purpose
A PWM drive switches three output legs between two DC rails. In a 60 Hz sinusoidal world the three phase voltages sum to zero at every instant, and that is the world the GFCI was designed around, but a drive does not live in it. The instantaneous sum of a drive’s three output voltages is never zero and cannot be, and that non zero sum is common mode voltage, stepping up and down at every switching event with edges in the tens to hundreds of nanoseconds.
A stepping voltage across a capacitance produces current, and with edges that steep it does not take much capacitance to produce a meaningful one. The system is full of it. There is capacitance from phase conductor to shield along the whole length of the motor cable, from winding to frame inside the motor, from rotor to stator across the air gap. The largest and most deliberate one sits inside the drive itself, where the EMC filter has capacitors wired from each phase to ground for the express purpose of giving that current a short path home.
Which means the drive’s filter was engineered to put high frequency current onto the grounding conductor, and the GFCI was engineered to open when current appears on the grounding conductor. Both devices are doing exactly what they were designed to do, and they are doing it against each other.
It is also not one frequency. Doepke, which has been building residual current devices for this problem longer than most of the industry has been worrying about it, lists five separate contributors coming off a converter: smooth DC from the six pulse bridge under fault conditions, a low order ripple from the DC link commutation, components at the switching frequency itself in the 2 to 16 kHz range, harmonics of that switching frequency, and components at the motor output frequency. Five contributors spread across five parts of the spectrum, which is why a single threshold measured at one frequency was never going to hold. KEB puts the appliance end of that band at up to 150 kHz, against a Class A device qualified at 60 Hz.
This is the same current that the VFD Installation Guide is about from end to end, and it is the same current that ends up in motor bearings. A GFCI complaint and a bearing complaint on the same machine are usually one phenomenon with two different exits.
What the HF rating actually certifies
In August of 2025 UL added Supplement SB to UL 943, an optional high frequency rating. A GFCI marked HF has been evaluated for immunity to differential currents below 4 times its pickup value across the band from 60 Hz up to 150 kHz, and immunity is the whole of what the mark claims. The device has been tested to confirm that it will not trip on that current.
Detection at those frequencies was supposed to be a separate device, the HF+, and the HF+ does not exist. There is nothing to specify and nothing to buy. Reviewing the November 2024 proposals to revise UL 943, CPSC staff said it supported folding the HF requirements into the mandatory Class A requirements rather than leaving them optional, and supported the concept of HF+ as well, but added that more research was necessary before a trip level could be established. Until that research lands, the code can only point at the half of the answer that exists.
Raising the threshold as frequency climbs is not a safety concession, which is the part worth having straight before somebody accuses you of one. IEC 60479 is where the 30 milliamp figure comes from in the first place, and the same work shows that the current required to put a heart into fibrillation rises substantially above power frequency. Schneider Electric states it plainly in its own guidance: the human body is less sensitive to high frequency current. Doepke goes a step further and names the reason, which is that the dominant hazard shifts as frequency climbs, from cardiac fibrillation near 50 and 60 Hz to thermal and electrochemical effects higher up. A device that holds a flat 5 milliamp threshold from 60 Hz to 150 kHz is not being more careful. It is applying a number derived for one frequency to a band where the physiology changed.
Europe built the other device, and drive makers tell you to avoid it
Here is where the American conversation is about to repeat an argument that is already settled somewhere else.
The European residual current device is typed rather than marked, and Type B is the one specified for three phase drives. ABB publishes the trip band as a multiple of the device’s rated residual current, and it is a graded curve rather than an on or off tolerance:
| Frequency | Tripping current, as a multiple of rated |
|---|---|
| 50/60 Hz | 0.5 to 1.0 |
| 150 Hz | 0.5 to 2.4 |
| 400 Hz | 0.5 to 6 |
| 1000 Hz | 0.5 to 14 |
The lower bound holds at half the rated current across the whole range, so the device never goes blind. The upper bound is what buys the tolerance, and by 1 kHz the device is permitted to let 14 times its rated residual current pass before it opens. Doepke publishes the same 14 times figure at the top of the Type B range. That table was in print roughly 10 years before UL wrote Supplement SB, and it is the shape of the answer the American device is reaching toward.
Then there is Type B+, which is the German variant and the closest thing in service to what HF+ is supposed to become. It has to detect residual current all the way up to 20 kHz, with the tripping current capped at 420 milliamps at any frequency in that range. It keeps its sensitivity exactly where a drive makes its leakage, which is the whole point of the design.
Lenze tells its customers not to use it. The wording in Lenze’s own application guidance is that their controllers usually cannot be operated together with Type B+ devices, because Type B+ holds its threshold as low as the rated current even in the upper frequency range while Type B is deliberately less sensitive up there. Schneider describes Type B+ as prone to nuisance tripping with switching mode converters, in a document whose purpose is selling residual current protection.
So the device that solves the detection problem is the device that reintroduces the tripping problem, and that has been true in a market that already shipped both for about a decade. The HF+ is not late only because a standards committee has been slow. It is late because the two goals pull against each other, and nobody has built the thing that satisfies both.
The HVAC piece is still being fought over
210.8 also widened the HVAC net, raising the covered threshold to 60 A where it used to stop at 50 A and below, with an exception that permits special purpose ground fault protection, a different device class with a higher trip threshold aimed at equipment rather than at personnel.
The separate exception for listed outdoor HVAC equipment in 210.8(F) has already run out. It expired September 1, 2026. AHRI filed a tentative interim amendment seeking to extend it to September 1, 2028, ACCA supports it, the NFPA comment period closed on September 9, and the outcome is still pending. Adoption is nowhere near uniform either: roughly 19 states had already amended or limited the requirement, and several are still enforcing pre-2020 editions that do not contain the section at all. If you are quoting rooftop units or condensing units this year, that is a phone call to your jurisdiction rather than a guess.
Normal does not mean fixed
Drive leakage current being normal does not make the amount of it a given. The current has to get back to the drive and that part is not up for negotiation, but the route is decided by the installation, and two decisions move it more than anything else.
The first is the shield. A shielded motor cable actually carries more conductor to ground capacitance than unshielded building wire does, so it generates more common mode current, not less. What it buys you is where that current goes. Bonded 360 degrees at the drive gland and again at the motor, the shield gives the current a return path that closes locally between the drive and the machine, downstream of anything watching at the panel. Pigtail that shield, or coil it up and leave it in the box, and you have put an inductor in series with the only good path home. A couple of inches of pigtail is tens of nanohenries, which is nothing at all at 60 Hz and a real impedance at the knee frequency of a 100 nanosecond edge. The current still goes home either way. It goes through the conduit system, through the building steel, and through the equipment grounding conductor, which is precisely the path the protective device is watching. The same reasoning covers every other quiet break in a ground path, which is the argument in Painted Back Panels: The Ground Path That Isn’t There and in A Grounding Ring Cannot Outrun a Bad Install.
The second decision is the filter, and it runs the opposite direction from what most people expect. KEB makes the point in its own technical paper on this subject: lowering the impedance from phase to ground is exactly what an EMC filter does, and lowering it raises the leakage current coming in from the supply side, through the protective device, during normal operation. The filter that fixes your radiated emissions problem makes your tripping problem worse. Worse still is the common layout where a machine builder puts a filter on every drive in the cabinet, because the leakage adds up, and both KEB and Doepke say the same thing about the fix. One larger shared filter ahead of a group of drives produces less total leakage than a filter on each of them. Doepke adds that simultaneous startup compounds it and recommends staggering the starts.
Which puts the nuisance trip in an uncomfortable position. On a badly terminated run it may be the only instrument in the building telling you the shield is not doing its job. Put in an HF device and the trip stops. The current does not stop, and where it goes instead is bearings, encoder cables, and 4 to 20 milliamp loops, none of which announce themselves for months. A device that grants immunity removes the complaint while leaving the condition in place, and the condition carries a longer and more expensive bill than a tripped breaker ever did.
What to check before you change the device
Start at the termination, because that is where the money already went. The shield gets bonded 360 degrees at both ends, since the motor cable shield is a return conductor and a return conductor has to complete a circuit, and that rule runs opposite to the one for control cable shields, which land at one end only. Mixing those two up is the most common shield error in the trade and it is made by careful people.
Then confirm ground continuity from the drive enclosure to the motor frame, under 1 ohm. It takes less than a minute and it tells you whether the cable anybody paid for is doing anything at all. The rest of that sequence is in the VFD Pre-Startup Checklist, and it is worth running in order rather than by instinct.
Know what system grounding you are sitting on and what the drive’s EMC filter link or screw is set to, because on an ungrounded or high resistance grounded system a drive’s internal filter left connected becomes its own problem, and every manufacturer’s manual I have worked out of says so in writing.
Then count the drives on the circuit, because this is where the European practice has a number and American practice does not. The German installation rule, DIN VDE 0100-530, requires that the system be designed so operating leakage current stays under 0.4 times the rated residual current of the protective device. Lenze states its own version as one third, or put differently, pick a device rated three times the leakage you expect while the drives are running. Call it somewhere between 2.5 and 3 times headroom between what the installation normally leaks and what the device is rated to catch. That is a design rule rather than a code requirement on this side of the Atlantic, and there is nothing in Article 210 that asks you to calculate it, but it is the number that tells you whether you have a device selection problem or a circuit loading problem.
After that, look at what the drive itself is reporting, since a panel device and a drive are two independent ground fault detectors watching one circuit through completely different physics. VFD Ground Fault Diagnosis covers what the drive side of that is telling you.
If the terminations are right, the continuity is right, the filter configuration matches the system, and the circuit is not carrying more drives than the device was ever sized for, then it has become a device selection question and the HF rating is the answer the code now points you toward. That ordering is the same ordering the VFD Troubleshooting Guide uses everywhere else. Confirm the installation, then replace the component.
The half that is still unwritten
All of this sits in Chapter 2, which covers branch circuits and services, and the equipment driving the change is pool pumps, HVAC units, and EV chargers. The drive specific article is 430, in Chapter 4, and that coverage should land shortly.
What I want to know is whether the same acknowledgment carries onto the industrial side. A 200 hp drive on a long shielded run produces far more common mode current than a pool pump ever will, and ground fault detection there is mostly handled inside the drive rather than by a device at the panel, so the two protection schemes have never been asked to coexist in writing. Until they are, the plant is left reconciling them by hand.
The useful version of this change is not the breaker, and it is not a requirement, because there is not one. It is that the code finally described the current as something a drive makes rather than something an installer caused, and the next step is treating it as a design quantity with a route and a budget, the way DIN VDE 0100-530 already does. Deciding that route is the whole of what the VFD Installation Guide and Before the First Fault are for, and if the people landing those terminations on your site learned the trade before shielded VFD cable was common, that gap is what the training was built to close.
Frequently Asked Questions
Why does my GFCI trip when a VFD is running?
A variable frequency drive produces high frequency common mode current through its EMC filter capacitors and through the capacitance between the motor cable and ground. That current returns to the drive through the grounding conductor instead of through the conductors the GFCI is watching, so the device reads an imbalance and opens. It is normal drive behavior and not a fault.
What is a GFCI-HF and do I need one?
A GFCI-HF is a Class A GFCI carrying an optional high frequency rating that UL added to UL 943 in August 2025. It has been tested for immunity to high frequency differential current from 60 Hz up to 150 kHz, so it tolerates ordinary drive leakage without opening. The 2026 NEC does not require one. Section 210.8 gained an informational note pointing at the HF marking, which tells you the device exists and is recognized rather than mandating it.
What is the difference between GFCI-HF and GFCI-HF+?
HF means the device tolerates high frequency leakage current without tripping. HF+ was meant to go further and actually detect faults at those frequencies. CPSC staff supported folding both into the mandatory Class A requirements but said more research was needed before an HF+ trip level could be set. Europe's equivalent device, the Type B+ residual current device, detects up to 20 kHz and is sensitive enough that Lenze tells customers its controllers usually cannot run with one, which is a fair preview of the difficulty.
Is raising the trip threshold at high frequency a safety compromise?
No, and IEC 60479 is the reason. The 30 milliamp figure everyone quotes was derived at power frequency, and the same work shows the current needed to cause fibrillation rises substantially as frequency climbs, with the dominant hazard shifting toward thermal and electrochemical effects. A graded threshold tracks the physiology rather than ignoring it.
Will a high frequency GFCI fix nuisance tripping on a drive circuit?
It will stop the tripping. Whether it fixes anything depends on why the current was reaching the building ground in the first place, because a poor shield termination keeps pushing that current through bearings and signal wiring long after the breaker has gone quiet. Check the installation before changing the device.
Does shielded VFD cable reduce GFCI tripping?
It changes where the current goes rather than how much of it exists. Shielded cable actually carries more capacitance to ground than unshielded wire, so it generates more common mode current, but a shield bonded 360 degrees at both the drive and the motor closes that current in a local loop the panel device never sees. A pigtailed or unterminated shield gives that benefit away and routes the current through the building ground instead.
How much leakage current is too much for one protective device?
European practice puts a number on it where American practice does not. DIN VDE 0100-530 requires the installation be designed so operating leakage stays under 0.4 times the rated residual current of the device, and Lenze recommends one third, meaning a device rated about three times the leakage you expect in normal running. Adding drives to a circuit, or adding an EMC filter to each one instead of sharing a larger filter, is what pushes an installation past that line.
Does the 2026 NEC require GFCI protection on HVAC circuits above 50 amps?
The 2026 NEC 210.8 raised the covered threshold to 60 amperes, with an exception permitting special purpose ground fault protection on HVAC equipment. The separate exception for listed outdoor HVAC equipment in 210.8(F) expired September 1, 2026, and AHRI has a tentative interim amendment pending that would extend it to 2028. Adoption is uneven by state, so confirm where your jurisdiction stands before quoting a job.
The annual report: The State of the VFD 2026
Nineteen drive brands assessed, thirteen scored on what they will actually put on the public record. Free, in a Plant Edition for specification work and a Channel Edition for quoting.