Apple warns trade secrets fed to AI are irretrievable.

Apple has alerted legal and technology stakeholders to a novel trade secret threat emerging from artificial intelligence integration. In a supplemental brief filed Monday in support of expedited discovery in its ongoing litigation against OpenAI, the company warned that confidential data fed into AI agents or models could trigger irreversible and continually propagating misuse. The filing centers on allegations involving a former employee who allegedly utilized an AI system to run simulations and learn from proprietary information while working for OpenAI. Apple legal counsel emphasized that this scenario extends beyond traditional document theft, creating unique challenges in recovering control once sensitive data has been absorbed by a machine learning system. The situation underscores a growing industry-wide vulnerability. Experts note that while employee mobility and knowledge transfer are long-standing risks, AI introduces a structural complication. Legal scholar Camilla Hrdy of Rutgers University observed that employees now effectively carry corporate knowledge into external AI systems, drastically complicating data containment. Although existing legal remedies such as injunctions to cease usage, disclosure bans, and financial damages remain applicable, enforcing them in an AI context presents technical hurdles. Determining whether confidential data was merely referenced by a retrieval system or fundamentally integrated into a model weights requires precise technical assessment. Addressing these challenges has spurred discussion around machine unlearning, a process designed to remove the influence of specific datasets from trained models. Computer science professor Sijia Liu of Michigan State University noted that while deleting source files from a retrieval database is straightforward, excising information embedded during training is resource-intensive and technically complex. As a more immediate containment strategy, Liu suggested implementing detection systems that flag sensitive inputs or inter-agent data transfers, triggering automated access restrictions. Apple filing did not specify whether the ex employee AI interactions involved model training or simple retrieval, leaving the exact technical footprint under investigation. This legal maneuver mirrors broader tensions within the artificial intelligence sector. Earlier this year, xAI filed a separate lawsuit alleging that OpenAI facilitated the theft of proprietary code through a former engineer personal cloud and ChatGPT integration, though that case was dismissed in June. As litigation progresses, the Apple OpenAI dispute highlights a critical inflection point for corporate data security. The technology sector now faces mounting pressure to develop standardized protocols for AI data hygiene, machine unlearning, and cross organizational information governance. Until technical and legal frameworks mature to address AI specific data leakage, companies will likely continue relying on expedited discovery and targeted injunctions to mitigate irreversible information exposure.