News and Analysis of Artificial Intelligence Technology Legal Issues

Brainwaves, Plagiarism, Manufactured Evidence, and AI Source Code Theft: A Trade Secrets Story

Brainwaves, Plagiarism, Manufactured Evidence, and AI Source Code Theft: A Trade Secrets Story

Aside from a few criminal cases involving national technology secrets, trade secret lawsuits, by their nature, don’t often grab headlines. Mostly they involve unremarkable factual and legal issues involving misappropriation of company proprietary information.

Not so the case of Brainwave v. Arshee, where the facts and bold allegations by some of the parties offer a compelling story of alleged software theft, plagiarism, and manufactured evidence involving former high-ranking executives. Brainwave Sci., Inc. v. Arshee, Inc., slip. op. No. 21-cv-4402 (BMC) (E.D.N.Y. Dec. 13, 2021).  Its jurisprudential precedent isn’t to be ignored either: a company that invests in protecting its trade secrets–even ones involving open source software–can expect equitable relief in the form of a preliminary injunction to prevent those who steal company secrets from using the information. This post explores what happened in the Brainwave matter, the parties’ allegations, and the court’s reasons for granting plaintiff’s requested relief. As this is a blog about artificial intelligence, a summary of the machine learning technology at the heart of the case is first presented.

The trade secret technology

Plaintiff Brainwave Sciences, Inc. (BWS) markets an AI-powered software/hardware system called iCognitive® for use by law enforcement and the military to conduct interrogations. BWS’ website described the system as one that assesses brainwave signals from a head-worn EEG (electroencephalogram) sensor apparatus, and includes a display interface for its operators to make inputs and monitor the system.  The technology recognizes specific “P300” brain responses of a person who has a memory of the details of a crime stored in their brain. In a nutshell, the system (a) prompts a user, (b) displays stimuli, (c) captures appropriate “P300 brainwave responses,” (d) filters out noisy signals (false positives/negative), (e) detects attempted countermeasures by users, (f) analyzes signal data using statistical and machine learning algorithms, and (g) generates a report “for use by examiners, investigating agencies and legal authorities.”  The signal analysis component apparently includes a trained machine learning model that looks for patterns of semantic memory recognition reflecting information that may be stored in the brain. Notably, the company’s software comprises much open-source code stitched together, something software developers routinely do in the AI industry (because, why write code for routine operations that others have already created and provide online without restrictions for free?). Apparently, the iCognitive® code was stored on one of the big tech company’s cloud DevOps platforms.

The parties, the lawsuit, and “plagiarism”

Defendants include Dr. Lawrence Farwell, who was the Chief Science Officer in BWS’ predecessor company, and Dr. Thierry Maison, who was BWS’ Chief Technology Officer.  Both former executives reportedly had extensive access and input into the development and structure of what became iCognative®. As discussed by the court, Dr. Maison reportedly led BWS’ efforts in refining algorithms, source code, “back end” programming, and user interfaces.  Each had signed company agreements related to protecting BWS’ intellectual property rights, including trade secrets, when they on-boarded.

BWS’ lawsuit arose following reports from a sales partner that defendants’ brainwave monitoring product included a user interface that was identical to the interface designed and implemented by Dr. Maison during his employment with BWS.  Using Code Similarity Checker, Codequiry®, a “Measure Software Similarity” analysis was conducted, which found a 98-99% likelihood of “plagiarism” in defendants’ software compared to BWS’.

The trade secret case

BWS sought a preliminary injunction to prevent defendants’ use and disclosure of its trade secrets. As such, it had the burden to demonstrate, among other things, “a likelihood of success on the merits” of its federal Defends Trade Secrets Act (DTSA) claim. To meet that burden, BWS had to, (1) show that it owned a valid trade secret (one related to a product or service used in, or intended for use in, interstate or foreign commerce), and (2) that defendants created their program by stealing BWS’ trade secrets. 18 U.S.C. § 1836(b)(1).  Under the DTSA, a trade secret includes “all forms and types of financial, business, scientific, technical, economic, or engineering information…if – (A) the owner thereof has taken reasonable measures to keep such information secret; and the information derives independent economic value . . . from not being generally known . . .” 18 U.S.C. § 1839(3).

On the question of whether BWS employed “reasonable measures,” the court found that BWS went to “significant lengths” to protect its proprietary work, including measures often seen used in the software industry: securely storing pertinent source code, implementing physical office access security, requiring third parties to execute confidentiality and non-disclosure agreements prior to disclosing trade secrets to them, and requiring employees sign appropriately-tailored agreements covering proprietary and secret information.  The court dismissed for lack of evidence defendants’ testimony that BWS failed to protect its software because portions of BWS code were found by defendants online using a simple Google search. Relatedly, the court concluded that plaintiffs would likely be able to rebut defendants’ argument that defendants’ software was based solely on publicly-available information.

Indeed, the court was unpersuaded by defendants’ argument that BWS’ system was “merely” an amalgamation of widely known ideas, technology, and approximately 20 to 25 “pieces” of open-source code. “Such a piecemeal element- by-element focus obscures the forest for the trees,” it wrote. ‘If two houses have identical rooms with many identical features and floorplans, that is strong circumstantial evidence that one house was derived from the other — particularly where the architect of the later-built house had resided in the earlier one.” (citing Broker Genius, Inc. v. Volpone, 313 F. Supp. 3d 484 , 508 (S.D.N.Y. 2018)).

On the question of “economic value,” the Court relied on defendants’ admission that Dr. Maison chose to copy all the BWS code instead of finding its various pieces online and tying them together again. Dr. Maison testified that the former approach saved him “a lot of time.” As the court explained, the very fact that it took years for them to figure out how to uniquely combine open source code while working at BWS was persuasive evidence of final products’ “independent economic value…from not being generally known.” The court found it convincing that the system the defendants created was almost identical to BWS’, which is evidence that another approach was not apparent, even to two experts in the field.

Responding to the 98-99% likelihood of “plagiarism” evidence proffered by BWS, defendants suggested that plaintiff had re-engineered defendants’ code to facilitate its lawsuit. Specifically, Dr. Farwell and Dr. Maison alleged that, after they provided BWS’ counsel with their new program, BWS manipulated their code, submitted the altered code and Dr. Maison’s work to a third-party software evaluator, and then submitted that manufactured result as evidence to support its motion.  The court noted that defendants presented no meaningful evidence to support these “bold claims,” but nonetheless acknowledged it would be a valid line of inquiry to be pursued during discovery.  At the same time, the court faulted defendants for not presenting an alternative comparison analysis that was available to them. Specifically, defendants could have collected the relevant code claimed to be openly available online, documented its location, and compared that code with their new system to show that it too generated a 99% probability of plagiarism.

In addition to finding a “likelihood of success on the merits,” the court found in plaintiff’s favor on the “irreparable injury,” “balance of hardships,” and “public interest” elements needed for issuance of an injunction.

Some takeaways

In addition to underscoring the axiom that copying others’ work can lead to trouble, the Brainwave decision highlights an important trade secret principle: there is no presumption that using a significant percent of open source code in one’s software stack disqualifies it from trade secret protection.  Rather, a software company should assume valuable trade secrets may lie in the unique and proprietary way in which its software developers and engineers bring pieces of code together to create a commercially-viable technology, regardless of whether the pieces are individual snippets of machine learning code or entire programs available from open sources like Github, Stack, manuscript pre-print websites, and even code attached to papers submitted to tech conferences (though, copyright and other issues need to be considered). Less comforting is the knowledge that, as Brainwave suggests, even the most robust measures to protect company secrets may not stop employee theft of secrets. The best laid plans are not a panacea when it comes to trade secrets.

As noted elsewhere on this blog, trade secrets are not the only way a company can protect its intellectual property, proprietary information, and know how.  Software companies routinely also take advantage of U.S. patent and trademark laws. The often-asked question is whether a company should bother with patents (which require public disclosure of technology) or just rely on trade secret law.  In Brainwave, plaintiff averred that brainwave testing was developed by others and is the subject of now-expired patents, so the patents versus trade secrets debate was not an issue. While there is often room in a company’s intellectual property strategy for both patents and trade secrets, I lean in favor of protecting company secrets with patents, if available, especially in areas of technology where there is high employee turnover. As the Brainwave court and many others have said, “A trade secret once lost is, of course, lost forever.” FMC Corp. v. Taiwan Tainan Giant Indus. Co., 730 F.2d 61, 63 (2d Cir. 1984).

Software companies should also consider using copyright law to prevent others from using its software without permission, though the company must take steps to have its source code registered with the Copyright Office.  This option is available even where most of the software consists of preexisting works of others (which are disclaimed on the copyright application). Notably, a copyright registration of a “computer program” can extend copyright protection not only to the literal elements of the program—its source codes—but also to the screen displays it generates to the extent they contain original creative authorship. See Copyright Office Circular 61 (Copyright Registration of Computer Programs).

Finally, while not directly at issue, companies should note that the customary practice of protecting trade secrets by contract using employee non-competes and confidentiality agreements is being questioned by courts across the country, though narrowly-drafted employee agreements focusing on protecting an employer’s trade secrets and other confidential business information from being improperly used or disclosed have been upheld.