Patricia Thaine
Co-founder and chairwoman, Limina (formerly Private AI)
Research in privacy-preserving NLP
Patricia Thaine holds a master's degree in computer science from the University of Toronto[1]. A 2023 Techfinitive interview described her as a computer science PhD candidate at the university, on leave, and as an alumna of the Vector Institute[2]. Her research and development work has focused on privacy-preserving natural language processing[2].
With Gerald Penn she published papers on computing over encrypted data, including a 2019 Interspeech paper describing how to extract Mel-frequency and Bark-frequency cepstral coefficients from an encrypted speech signal without decrypting intermediate values[3]. The paper presented this as a first step towards secure end-to-end speech recognition over encrypted data[3]. At EACL 2021 Thaine and Penn presented a method that uses the Chinese Remainder Theorem to make word embeddings compact, task-precise and secure[4]. In 2020 they published Reasoning about unstructured data de-identification in the Journal of Data Protection & Privacy, which frames the de-identification of unstructured text within the wider field of privacy-enhancing technologies[9].
Founding Private AI
Thaine co-founded Private AI with Pieter Luitjens, who became chief technology officer, and Gerald Penn, the company's former chief science officer[1]. The company began in 2017 with an idea for searching encrypted data and pivoted by 2019 to give developers an easy way to build privacy into their software[1]. Thaine has said the company had its first customer in October 2020[2].
Private AI used AI to detect, redact and replace personal data in text, images, documents, audio and video[1]. In November 2022 it closed a Series A round of $10.7 million CAD ($8 million USD) led by BDC Capital, with Microsoft's venture fund M12 among the existing investors[1]. Thaine, then chief executive, told BetaKit the funding would support product development and expansion into Europe[1].
In the Techfinitive interview she said the identification of personal information was not being addressed for the 80 to 90% of collected data that is unstructured[2].
Limina and healthcare data
Private AI has since been renamed Limina, a name the company says comes from the Latin limen, meaning threshold[5]. Limina states that the product, customers and deployment options did not change with the new name[5]. The software removes personal data from text, images, audio and documents, and runs inside a customer's own cloud or on-premises environment[5]. Thaine is now the company's co-founder and chairwoman[5][6].
In an April 2026 HLTH podcast she discussed the privacy of unstructured data as a barrier to AI adoption in healthcare[6]. The episode covered how quasi-identifiers such as age ranges, locations and diagnoses add to re-identification risk, and how training models on real patient data can lead them to repeat that data in production[6]. She also explained that the company rebranded because the old name drew requests for on-premises language models, which it does not provide[6].
HLTH's speaker bio states that Private AI was named a 2023 Technology Pioneer by the World Economic Forum and that Thaine hosts The Data Frontier podcast[6].
Writing on PII minimization
Thaine co-wrote a chapter with Ken Huang and Yale Li in the 2024 Springer book Generative AI Security, titled Use GenAI Tools to Boost Your Security Posture[7]. In 2026 Apress published the PII Minimization Handbook, to which she contributed the chapters Minimization Techniques for Unstructured Data and PII Minimization in Healthcare[8].
Known for
- Co-founding Private AI, now Limina
- De-identifying unstructured data before it is used in AI
- Research on privacy-preserving natural language processing
Sources
- BetaKit: Private AI secures $10.7 million CAD, December 2022 — checked 6 Nov 2026
- Techfinitive: Patricia Thaine, Private AI, interview, 2023 — checked 6 Nov 2026
- ISCA Archive: Extracting Mel-Frequency and Bark-Frequency Cepstral Coefficients from Encrypted Signals, Interspeech 2019 — checked 6 Nov 2026
- ACL Anthology: The Chinese Remainder Theorem for Compact, Task-Precise, Efficient and Secure Word Embeddings, EACL 2021 — checked 6 Nov 2026
- Limina: Private AI is now Limina — checked 6 Nov 2026
- HLTH podcast: How Limina Is Making Sensitive Health Data Safe for AI, April 2026 — checked 6 Nov 2026
- Springer, Generative AI Security: Use GenAI Tools to Boost Your Security Posture, 2024 — checked 6 Nov 2026
- Apress, PII Minimization Handbook: PII Minimization in Healthcare, 2026 — checked 6 Nov 2026
- Journal of Data Protection & Privacy (Henry Stewart): Reasoning about unstructured data de-identification, 2020 — checked 6 Nov 2026