Imagine That IP Law Blog

Introduction
The recent debate over artificial intelligence and inventorship has often focused on a single question: can an AI system such as ChatGPT, DABUS, or another generative model be named as an inventor on a United States patent application? The Federal Circuit answered that question in Thaler v. Vidal, holding that the Patent Act limits inventorship to natural persons. Thaler v. Vidal, 43 F.4th 1207, 1211-13 (Fed. Cir. 2022). Yet the next wave of disputes is unlikely to end with Thaler. Parties will continue to argue that AI systems do more than merely assist human inventors, and some will press the position that AI can be a rightful joint inventor.

If that premise were accepted, however, the consequence would not be a broader, more inclusive inventorship doctrine. The consequence would be invalidity. Fortress Iron, LP v. Digger Specialties, Inc. makes the point sharply. There, the Federal Circuit reaffirmed that a patent must accurately identify all inventors, that omission of a true inventor is an error with invalidating force, and that the patent survives only if the error can be corrected under 35 U.S.C. § 256. If the error cannot be corrected, the patent is invalid. Fortress Iron, LP v. Digger Specialties, Inc., No. 2024-2313, 2026 U.S. App. LEXIS 9509, at *5-6, *10-13 (Fed. Cir. Apr. 2, 2026).

That logic has direct implications for AI. Suppose, contrary to Thaler, that an AI system such as ChatGPT can be rightfully named as a joint inventor because it made a significant contribution to conception. In that event, any patent that omitted the AI from the inventor list would suffer from nonjoinder of an actual inventor. Under Fortress Iron, omission of a true inventor invalidates the patent unless the omission can be corrected under § 256. Id. at *10-13. But an AI cannot execute the statutory inventor oath or declaration in the manner contemplated by the Patent Act, and under current law it is not an “individual” at all. Thaler, 43 F.4th at 1210-13. Thus, even on the counterfactual assumption that AI could be a rightful joint inventor, the result would not be a valid patent with expanded inventorship. It would be a patent rendered invalid by an uncorrectable inventorship defect.

This conclusion exposes an important structural point. The current statutory regime does not merely decline to recognize AI inventorship. It makes recognition of AI inventorship incompatible with patent validity. Once Fortress Iron is read together with Thaler, the doctrinal picture becomes clear: if AI must be listed, the patent fails; if AI cannot be listed, then AI cannot be a lawful inventor in a way that preserves patent validity. The system leaves only one coherent path, namely, to treat AI as a tool used by human inventors rather than as an inventor itself.

I. Fortress Iron Restates the Centrality of Correct Inventorship

Fortress Iron arose from a straightforward inventorship dispute. The patents at issue omitted an agreed coinventor, Huang. The patentee was able to add another omitted coinventor, Lin, through the administrative mechanism of § 256(a), but could not locate Huang and therefore could not provide him notice and an opportunity to be heard. Fortress Iron, 2026 U.S. App. LEXIS 9509, at *1-4. The Federal Circuit affirmed summary judgment of invalidity.

The court began with first principles. “A patent must accurately name those who invented its claimed subject matter.” Id. at *5. Failure to do so renders the patent invalid, although Congress has provided a savings mechanism in § 256 for patents whose inventorship errors can be corrected. Id. at 5-6; see also Pannu v. Iolab Corp., 155 F.3d 1344, 1349-50 (Fed. Cir. 1998). The court described § 256 as a “savings provision,” but only to the extent its statutory requirements are actually satisfied. Fortress Iron, 2026 U.S. App. LEXIS 9509, at *5-6, *9-10.

That qualification did the decisive work. Section 256(b) allows a court to order correction only “on notice and hearing of all parties concerned.” 35 U.S.C. § 256(b). The omitted inventor, Huang, was a “party concerned,” and because Fortress could not locate him, it could not satisfy the statutory prerequisite. Fortress Iron, 2026 U.S. App. LEXIS 9509, at *6-10. The result followed inescapably: because the patent omitted a true inventor and the omission could not be corrected according to law, the patent was invalid. Id. at *10-13.

The Federal Circuit then made the broader doctrinal point explicit. Section 256(b) says that omission of inventors “shall not invalidate” the patent “if it can be corrected.” 35 U.S.C. § 256(b). The “necessary and opposite implication,” the court explained, is that a patent is invalid for omitting inventors when the error cannot be corrected. Fortress Iron, 2026 U.S. App. LEXIS 9509, at *10. Fortress’s attempt to argue that it was enough for the patent to list at least one true inventor was rejected because it would render the statutory text meaningless. Id. at *10-12. When there are multiple inventors, all must be listed. Id. at *11-12.

That holding matters far beyond the particular facts of a missing human coinventor. It establishes a simple but powerful rule: proper inventorship is not merely a formal requirement of patent prosecution; it is a validity requirement. If a person or entity is a true inventor, that inventor must be named, and failure to name that inventor invalidates the patent unless the statutory correction mechanism is available.

II. Joint Inventorship Doctrine Makes the AI Question More Dangerous, Not Less

One might think that joint inventorship doctrine softens the stakes by allowing multiple contributors with unequal roles to share inventorship. In one sense, it does. The Federal Circuit has repeatedly explained that joint inventors need not contribute the same type or amount, need not work at the same time, and need not contribute to every claim. Dana-Farber Cancer Inst., Inc. v. Ono Pharm. Co., 964 F.3d 1365, 1371-73 (Fed. Cir. 2020). Dana-Farber emphasizes that a joint invention is the product of collaboration and that partial but significant contributions can suffice. Id. at 1371-73.

Likewise, Ethicon, Inc. v. United States Surgical Corp. reiterates that conception is the touchstone of inventorship, that one who merely explains well-known concepts or assists after conception is not a joint inventor, and that a contribution to even one claim can be enough. Ethicon, Inc. v. U.S. Surgical Corp., 135 F.3d 1456, 1460-61 (Fed. Cir. 1998). These rules make joint inventorship flexible. But they also make it potentially expansive. If an AI system were thought capable of contributing to conception in a legally meaningful way, then the doctrine of joint inventorship would not marginalize that contribution. It would elevate it.

That is where Fortress Iron becomes so important. The doctrinal flexibility of joint inventorship does not reduce the importance of correct naming. It heightens it. Once a contribution crosses the line from assistance to conception, nonjoinder becomes fatal unless corrected. Fortress Iron expressly relies on that premise, explaining that omission of an actual inventor invalidates the patent and that § 256 is only a conditional escape valve. Fortress Iron, 2026 U.S. App. LEXIS 9509, at *5-6, *10-13.

Thus, anyone arguing that ChatGPT can be “rightfully named as a joint inventor” is making a claim with consequences that extend beyond semantics. The argument is not just that AI may deserve credit. It is that AI’s omission from issued patents is a statutory inventorship error of the sort that Fortress Iron says invalidates a patent unless corrected.

III. Thaler Closes the Correction Path

If Fortress Iron supplies the invalidity rule, Thaler v. Vidal closes the escape route. In Thaler, the Federal Circuit held that the Patent Act unambiguously requires an inventor to be a natural person. Thaler, 43 F.4th at 1210-13. The court anchored that conclusion in the statutory definition of “inventor” as an “individual,” the related use of “individuals” in the joint inventorship provisions, and the oath and declaration requirements of § 115. Id. at 1210-12.

The court emphasized that the Patent Act uses personal pronouns, requires inventor oaths or declarations, and contains no indication that Congress intended “individual” to extend beyond human beings. Id. at 1210-12. It further noted that, although AI may be involved in the inventive process, the case before it did not concern inventions made by humans with AI assistance. Id. at 1213.

That holding means that if an AI were omitted from a patent because it cannot lawfully be listed, the patentee could not invoke § 256 to fix the problem by naming the AI. A correction under § 256 still requires a lawful inventor designation under the Patent Act. But Thaler says AI is not an inventor under that Act. Id. at 1210-13.

The result is a doctrinal trap for anyone who insists that AI can truly invent. If the AI is a true joint inventor, omission invalidates the patent under Fortress Iron unless correction is possible. Yet correction is impossible because the omitted “inventor” is not an “individual” under Thaler and cannot satisfy the statutory inventor requirements. The patent therefore remains invalid.

In other words, Fortress Iron and Thaler together create a closed system. Thaler bars AI from being named. Fortress Iron invalidates patents that fail to name all true inventors when the defect cannot be corrected. Put together, they foreclose the proposition that AI can be a rightful joint inventor while preserving patent validity.

IV. Why the Better View Is That AI Is a Tool, Not an Inventor

The strongest practical implication of this analysis is that the patent system has powerful reasons to resist reclassifying AI from tool to inventor. If AI were treated as an inventor, large numbers of patents could be cast into doubt. Applicants regularly use software tools, language models, data-analysis systems, and design optimization platforms in inventive work. If those systems can cross the line from assistance to conception, then inventorship litigation would shift toward identifying machine contributions and asking whether omission of the machine invalidates the patent.

That result would be deeply destabilizing. It would create uncertainty not only about future filings but also about past patents prosecuted on the assumption that only humans can invent. It would also complicate ownership. Ethicon explains that a coinventor presumptively owns a pro rata undivided interest in the entire patent. Ethicon, 135 F.3d at 1465-66. While that principle was developed for natural persons, it illustrates the broader structural mismatch between patent law and machine inventorship. Inventorship is not simply honorary credit. It carries legal, financial, and ownership consequences, a point Fortress Iron itself underscores. Fortress Iron, 2026 U.S. App. LEXIS 9509, at *8-10.

The current doctrine instead points toward a more stable line: AI may assist humans in generating, refining, testing, or expressing ideas, but the legal inventors remain the natural persons who form the relevant conception and take responsibility for the claimed subject matter. That approach is consistent with Thaler’s insistence on human inventors and avoids the invalidity trap that Fortress Iron would trigger if AI were deemed a true omitted inventor. Thaler, 43 F.4th at 1213.

V. Conclusion

Fortress Iron v. Digger Specialties should play a central role in the AI inventorship debate. The case confirms that correct inventorship is a validity requirement, not a mere procedural nicety, and that omission of a true inventor invalidates a patent unless the error can be corrected under § 256. Fortress Iron, 2026 U.S. App. LEXIS 9509, at *5-6, *10-13. When that rule is read together with Thaler v. Vidal, the consequences for AI are unavoidable. If an AI such as ChatGPT can be rightfully named as a joint inventor, then its omission from the patent is an inventorship defect. But because AI cannot lawfully be named as an inventor under the current Patent Act, the defect cannot be corrected. The patent is therefore invalid.

That conclusion does not merely rebut the case for AI inventorship. It shows why the current system cannot absorb AI inventorship without breaking. As long as the Patent Act defines inventors as human individuals, the only coherent path is to treat AI as a tool used by human inventors, not as an inventor itself. If that path is thought inadequate as a matter of policy, the remedy lies with Congress, not with an attempt to stretch existing inventorship doctrine into a form that Fortress Iron makes fatal to patent validity.