Ellison Anne Williams did not set out to encrypt the world. She arrived at college with medicine in mind, doubled up on mathematics and French, and found that the numbers were having more fun. “This math thing is kind of fun,” she later recalled. One degree led to another: a master’s in mathematics, a doctorate in algebraic combinatorics, then a second master’s in computer science with a machine-learning focus. By the time she entered the U.S. intelligence community in 2004, the detour had become a direction.
Her doctoral work had an austere title, A Formula for N-Row Macdonald Polynomials. Her government work was less publishable. Over twelve years, including time connected to the National Security Agency and Johns Hopkins University Applied Physics Laboratory, Williams worked on large-scale analytics, network modeling, distributed computing, and cryptographic applications. The recurring problem was coordination: hundreds or thousands of machines had to calculate together, and valuable information had to be used without being carelessly exposed.
There are three familiar conditions for data. It rests in a file. It travels through a network. Or it is being used. Security products became good at the first two. The third remained awkward because computation usually requires data to be visible somewhere. Williams built Enveil around that exposed instant.
The company inside the classified problem
Homomorphic encryption offers a peculiar promise: perform operations on encrypted information as if it were plain text. A database can return a useful result without seeing the search in readable form. The owner of the query keeps its intent. The owner of the data keeps its records. Neither side needs to perform the corporate ritual of copying everything into a new, tempting pile.
For years the idea carried a laboratory smell. It was computationally expensive, elegant on paper, and easy to admire from a safe distance. Williams’s team combined cryptography with distributed algorithms to make selected operations practical at scale. Her insight was partly technical and partly architectural: use the mathematics creatively, minimize what it must do, and fit the result around the systems customers already have.
We never expose the operation, the results, or the data itself.Ellison Anne Williams, at RSA Conference in 2017
The technology had developed inside a broader government effort. When an avenue opened to commercialize a piece of it, Williams took the exit. She had been waiting for one. Entrepreneurship ran through her family: her father and grandparents had built businesses, and she had run a cake business in high school. A government technologist dreaming of company formation can resemble a greyhound in a library, beautifully composed and plainly designed for another kind of room.
In 2016, she left after twelve years and founded Enveil near Fort Meade in Maryland. Several former colleagues joined. DataTribe, an incubator focused on technology emerging from government work, gave the new company a place to land. The name compressed “encrypted veil” into something that sounded like both a verb and a curtain.
Three minutes to explain the veil
Four months after Enveil’s founding, Williams stood at the 2017 RSA Conference Innovation Sandbox. More than 300 companies had applied. Ten finalists received three minutes each. This is not a generous format for algebra, distributed systems, and enterprise security, but it is an excellent test of whether an inventor knows which part matters.
Williams explained the API, the parallelism, and the ability to process data while it remained encrypted. Enveil reached the final two and placed second. The young company had technical evidence and, now, a public story. One could protect the question, the answer, and the data owner’s material while still completing useful work.
The category around Enveil widened. “Data in use” became part of a larger privacy-enhancing technologies vocabulary. The customers’ problems also became more legible. A bank wants to compare suspicious patterns with another institution but cannot simply swap customer files. A government team wants to search information across classifications or jurisdictions. An AI group wants a model to act on sensitive data without extracting it from the owner’s environment. The common shape is a boundary with value on both sides.
keeps its query private
COMPUTE
keeps its data private
That framing helped privacy shed its reputation as a department of “no.” Enveil sells an allowed action. Search over there. Compare across this wall. Train or evaluate without collecting the raw material. The security property matters because it creates permission for work that policy, ownership, competitive risk, or classification would otherwise stop.
The operator with a crowded timetable
Williams describes herself as hyper-efficient. A profile of her home and work life found five children, three rescue racing greyhounds, two bearded dragons, two chinchillas, a fellow mathematician for a spouse, plus swimming and cycling. Her summary was crisp: “I like to keep 1,000 trains running at the same time.” Investor Bob Ackerman compared working with her to being inside a Swiss watch.
The image is funny because clocks are orderly and startups are not. Enveil had to convert research into software, software into a buying category, and a buying category into contracts with institutions that do not enjoy improvising with sensitive data. Williams had spent years learning how large bureaucracies move. The very environment that felt ill-suited to her entrepreneurial temperament became preparation for enterprise sales, compliance, and mission work.
Her public career advice carries the same patience. Careers are nonlinear, she has said. Choose opportunities for what they will teach. Ask experienced people for help before tuition is collected by the school of hard knocks. A route that began in pre-med, wandered through French and algebraic combinatorics, and passed through classified systems has earned the right to distrust a neat five-year plan.
Careers are nonlinear.Williams's advice to early-career professionals
She has also spoken about substance and representation in security. At the OURSA conference, created in response to an all-male RSA keynote slate, Williams argued that women in the field should be present for their expertise, not as ceremonial balance. North Carolina State later honored her as an Outstanding Young Alumna. She mentors, speaks about women in STEM and mothers in leadership, and serves on the RSA Conference Program Committee.
The question travels farther
By 2025, the applications had reached financial crime and space. Williams wrote that institutions fighting fraud often know collaboration is necessary but remain trapped by fragmented ownership, legal boundaries, and the risk of moving sensitive records. Her proposed shift is from copying data toward computing across it. The distinction sounds grammatical until one imagines the breached warehouse that never had to exist.
In a conversation about civilian, commercial, and military space assets, she described privacy-enhancing technology as connective tissue for decentralized analysis. The use case surprised her. Had someone predicted “PETs in space” when she founded Enveil, she said she would have laughed. Yet the pattern was familiar: heterogeneous systems, valuable signals, and owners that cannot flatten every boundary merely because a shared answer would help.
Secure AI is the current extension. Models need rich, relevant data, while operational systems face higher stakes than sandboxes. Williams argues that protecting information during computation lets an organization use a model across sensitive environments without requiring every participant to abandon control. It does not resolve every question of AI governance. It addresses a precise one: how to calculate when visibility itself creates risk.
Enveil’s ZeroReveal software reached version 6.0 in 2026, with encrypted search still at the center and secure analytics, collaboration, and AI around it. The company continues to work in finance, public-sector missions, and cross-jurisdiction data use. The technical primitive has acquired an ecosystem of ordinary verbs: search, compare, detect, train, share. That is often how infrastructure wins. The specialist vocabulary disappears into an action someone needed to take before lunch.
Williams’s story is frequently described as a leap from the NSA to entrepreneurship. It looks more like a transfer of method. Pure mathematics taught her to stay with abstraction. Distributed computing taught her to split a hard task across many machines. Government taught her how boundaries behave. Company building required all three, then added the awkward human proof: explain why the protected question is worth asking.
The veil in Enveil is not a wall. It is a negotiating device. Each party keeps what must remain private and contributes just enough computation to produce an answer. In an economy that treated data collection as a reflex, Williams built for restraint with utility. The machines may see less. The people using them may learn more.