USA – MASSACHUSETTS Trends and Developments Contributed by: Adam Gershenson and Audrey Pope, Weil, Gotshal & Manges LLP
failed where the operative agreement between parties no longer imposed a surviving duty on the defendant to limit its use of the purportedly secret information. For companies deploying or integrating AI tools through vendors, pilots, or collaborators, one takeaway seems clear: reasonable measures are increasingly likely to be measured, in keeping with longstanding precedent, by examining both technical and contractual safeguards. This imposes an ongoing obligation to protect informa - tion with both well-defended cybersecurity architecture and contracts that ensure every relevant party in the chain is bound to protect the information after access is granted. Nor should companies overlook the risk of self-disclo - sure by their own employees and contractors, whether inadvertent or malicious. For example, if developers carelessly embedded sensitive information like esca - lation procedures in system prompt code, believing that proprietary infrastructure elements will be guard - ed from user view, that information, too, is vulnerable to exposure. The easier it becomes for employees and contractors to paste sensitive or proprietary infor - mation into public-facing AI tools, the harder it may become to later show that the company reasonably protected that information as secret. Acceptable-use policies, internal model-governance rules, training, and audit trails are both key hygiene issues for IT and compliance teams, and part of the ammunition the legal team will need to assert trade secret claims. Game changer – vibe coding In February 2025, OpenAI co-founder Andrej Karpa - thy posted on X about his nascent interest in “vibe coding”. The phrase refers to a method of software development that allows users to turn plain-language LLM prompts into code. While Karpathy himself is no computer science novice, the practice has made cod - ing and software development possible for anyone with access to a general-purpose chatbot. The accessibility of vibe coding lowers the cost of copying proprietary products in at least two related ways. First, vibe coding requires far less technical expertise than traditional coding. If a company wants to produce a product that is like something already
on the market, it may be able to compete by merely telling the model what it wants to create, or copy. This may enable a company to enjoy accelerated product development without getting embroiled in the increas - ingly competitive – and potentially litigious – talent wars. This will likely shift market dynamics and allow non-frontier companies to reap second-mover advan - tages without first-mover investments. Second, vibe coding works because the AI model can take what is readily ascertainable about a product and, on its own, generate information about the here - tofore “secret” elements of the product (or, at least, information about what those elements are likely to be). It facilitates essentially cost-free “independent development” or reverse engineering. Under existing standards, this would seem to undermine trade secret claims in the underlying material. It is possible that courts, as with system injection prompts, will reject the propriety of vibe coding when evaluating trade secret claims, but the conclusion is far from inevitable. Even if doctrinal developments do protect trade secret information as a matter of law, the existence of vibe coding will still have market effects, as copycat and alternative products are increasingly easily and cheap - ly produced. This suggests that whatever way the doctrine evolves, companies should invest deliberate - ly in a blend of overlapping protections, both legal and technical. As proprietary information becomes more vulnerable, companies may, for example, become more reliant on market-based protections like brand identity. Coca-Cola’s market dominance suggests that it may matter less than expected if someone can cre - ate a “chemically identical” soda – people will still feel drawn to its iconography, history, and associations. The brand’s 1942 slogan, “[t]he only thing like Coca- Cola is Coca-Cola itself”, may no longer be true, but customers may still think of it as the “real thing”, or long for “a Coke and a smile”. If this is the new real - ity for even the most well-known, most paradigmatic trade secret, companies in Massachusetts – and eve - rywhere – will need to use technical measures and branding devices to protect their secrets, both from potential theft and uncertainty in the law.
208 CHAMBERS.COM
Powered by FlippingBook