Somewhere in a filing cabinet at IBM in August 2015, five researchers wrote down an idea: put a number on how well the car’s computer is driving, put a number on how well the human is driving, compare the two, and hand the wheel to whichever one is winning.
That idea is now three U.S. patents, and it is being used to sue Tesla over Full Self-Driving. Tesla tried to have the Patent Office kill those patents. The Patent Office refused to even look. The Federal Circuit said no court has the power to make it look. And on July 27, Tesla asked the Supreme Court to decide whether that’s actually how the law works.
If you build, modify, or sell anything with a driver-assist system in it, this obscure administrative fight matters more to you than most product launches.
The patents
The asserted claims — 11,597,402, 11,738,765 and 12,037,004 — all descend from an August 2015 application assigned to International Business Machines. The specification describes an on-board control processor and a human driver, each scored for competence under current roadway conditions, with control assigned to whichever scores higher. Weather, sensor anomalies, driver distraction — all feed the comparison.
Related Articles
- Wichita’s License-Plate Cameras Misidentify Cars Up To 20% Of The Time. Now A Local Is Suing To Rip Out All 200
- Ford Recalls 86,543 Mustang Mach-Es Because Its Factory Got Faster Than the Glue Could Handle
Read that again and think about every Level 2 system currently on sale. Torque-sensing steering wheels, infrared driver-monitoring cameras, systems that nag you and then pull to the shoulder. The hand-off logic is the whole ballgame in supervised autonomy, and this family claims a fairly abstract version of it with a priority date that predates most of the hardware now shipping. The patents don’t expire until September 2035.
IBM didn’t keep them. The assignment record runs IBM to Daedalus Group, then to Slingshot IoT in January 2020, then to Granite Vehicle Ventures in July 2023 — an entity that builds nothing and sued Tesla seventeen months later in the Eastern District of Texas. This is the ordinary lifecycle of a monetized patent portfolio, and it’s why the administrative review process exists in the first place.
The trap
Congress created inter partes review in the America Invents Act so accused infringers could take a validity fight to expert patent examiners instead of a jury pool. It’s cheaper than litigation, though not cheap: 37 CFR 42.15 sets the request fee at $23,750 for up to 20 claims, with another $28,125 due on institution.
Tesla filed four IPR petitions in May 2025, comfortably inside the one-year window § 315(b) allows after service. It then did more than the statute requires, offering a stipulation surrendering invalidity defenses in court to eliminate any duplication concern.
Denied anyway. As the Federal Circuit later recorded, the agency reasoned “it is unlikely that a final written decision in [these] proceeding[s] will issue before the district court trial occurs,” and that review would result “in significant duplication of effort, additional expense for the parties, and a risk of inconsistent decisions.”
The trial that outran the Patent Office was Judge Rodney Gilstrap’s, with jury selection set for August 3, 2026. Here’s the part that should make anyone’s eye twitch: Gilstrap later sent the case west. It now sits in the Northern District of California as case 5:26-cv-01457, where a magistrate was still resolving discovery disputes in July 2026. The speedy Texas trial that justified the denial no longer exists. Tesla’s one-year IPR window does not reopen.
Why no court will touch it
The Federal Circuit denied Tesla’s mandamus petition on February 27 in a three-page nonprecedential order. Section 314(d) makes institution decisions “final and nonappealable.” Absent a colorable constitutional claim, the panel said, mandamus is unavailable. Tesla dressed its ultra vires argument as a separation-of-powers problem; the court cited Dalton v. Specter for the proposition that alleging an official exceeded statutory authority isn’t a constitutional claim.
Which produces the circularity at the heart of all four cert petitions now pending: the agency can invent a criterion Congress never wrote, and the only body that could say so has been told it isn’t allowed to ask.
How the rules changed
None of this happened by accident. In March 2025 then-Acting Director Coke Morgan Stewart split institution into two phases, pulling discretionary questions up to the Director’s office and inviting briefing on considerations found nowhere in the statute. In October 2025, Director John Squires went further, revoking the delegation and deciding institution personally, largely through summary notices. The same day, the Office published a proposed rule that would convert these practices into categorical bars.
Related Articles
- Waymo Puts Robotaxis Back On Freeways — Here’s What Actually Broke
- Ford Buried a Timing Belt in Motor Oil to Save Gas — Now the Feds Say 135,000 Cars Are a Safety Risk
The numbers moved accordingly. USPTO’s own FY25 statistics show institution rates that had sat in the mid-to-high sixties across FY22 through FY24. They are now roughly half that.
Notably, when Google took the same question up on the “settled expectations” theory — patents denied review for being old — the Solicitor General waived the right to respond on behalf of the USPTO in May. The patent owner filed its opposition in July, arguing the case is a poor vehicle.
What this means if you’re not Tesla
Tesla can afford district court. A Tier 2 supplier building radar brackets cannot, and neither can a small EV converter or a performance-electronics shop that gets a demand letter over telematics or ADAS calibration. IPR was the affordable escape hatch, and it has narrowed to a crack.
The fallback is ex parte reexamination under § 302, which never went away. The request fee is $6,300 for a large entity, $2,520 for a small one, $1,260 for a micro entity. The catch is that you hand the examiner your prior art and then sit down — no cross-examination, no depositions, no participation in the back-and-forth with the patent owner. It’s a blunt instrument compared to a full IPR, but it is currently the cheapest one still on the shelf.
The second-order effect is the one that reaches the showroom. Validity fights that used to cost a few hundred thousand dollars in front of examiners now cost millions in front of juries, and the supplier contracts that allocate indemnity for exactly this exposure get repriced accordingly. Litigation cost is a line item in a vehicle program like anything else.
The Court hasn’t said whether it will hear any of this. But four petitioners, one question, and an agency that changed a decade of practice through internal memoranda make for the kind of posture that occasionally gets four votes. Tesla’s argument boils down to a simple test: if the agency’s reasons are truly beyond review, nothing stops it from inventing arbitrary ones. That’s a question worth answering before the hand-off logic in your car becomes somebody’s licensing revenue.