A person who studies anonymity for a living learns to distrust a clean diagram. Draw a network and everyone looks equal: little circles, tidy lines, no faces. Put people on it and the questions begin. Who can see a message? Who can tell where it came from? What happens when an attacker watches long enough? George Danezis built much of his early career around those questions. They carried him from Cambridge research on anonymous communication to University College London, then into the noisier world of startups, Facebook’s abandoned crypto ambitions and the blockchain company Mysten Labs.
The route can look like a change of subject. Anonymity networks hide the path a message takes. Blockchains deliberately expose a record that many computers can inspect. Yet both ask how strangers can rely on a system when some participants may behave badly. Danezis has followed that problem across institutions and technologies, often arriving at the least glamorous part of the promise: what actually works when software is deployed.
The watcher in the room
At the University of Cambridge, Danezis wrote a PhD thesis called Better Anonymous Communications, completed in 2004 under security researcher Ross Anderson. The title has the spare confidence of a research programme. Its subject was difficult because secrecy on a network is about more than encrypting the words inside a message. Timing, volume and routes can betray people even when the contents stay unreadable. In 2005, Danezis and Steven Murdoch published work showing how traffic analysis could narrow the anonymity offered by Tor. The point was neither that privacy had failed nor that it was easy. The point was that it needed to survive observers who studied the traffic as carefully as its designers did.
That appetite for testing a claim against an adversary shaped years of work on anonymous communications and privacy engineering. He contributed to designs associated with Mixminion, Sphinx and Loopix. By 2017, when he delivered his inaugural lecture at UCL, the subject was still alive enough to fill a talk titled “Foiling online surveillance: new developments in anonymous communications and their applications.” A career in privacy can be measured in papers, but its real unit is the person who can communicate without being exposed merely by using the service.
A university provided room to examine many possible answers. At UCL, Danezis became a professor of security and privacy engineering, a role the department still lists on a part-time basis alongside his work at Mysten. The two jobs look different from outside. He has described the difference more precisely: research explores the range of what could be done; building a live system forces a choice about what people need. “There really isn’t much difference,” he said of the two activities in a 2023 interview. The distinction comes when an attractive possibility must become a usable feature.
“It’s not just intellectual interest anymore, but value for users.”George Danezis, discussing research and production
From a laboratory to a company
As Ethereum made programmable blockchains visible to a wider audience, Danezis became interested in execution: how smart contracts could run without a popular application making a network slow and expensive. He and collaborators at UCL started Chainspace to commercialize work on a scalable, sharded smart-contract platform. Sharding divides work among parts of a system; the hard part is keeping the whole thing trustworthy while those parts operate. The Chainspace team was acquired by Facebook in 2019, taking its research into the effort that became Libra and later Diem.
There is an appealingly awkward turn in that move. The scholar of privacy and decentralization joined a company whose scale already covered much of the world. The work there concerned a proposed digital currency and the infrastructure around it. It was an opportunity to solve a distributed-systems problem at enormous scale, and it also placed the problem inside a corporate project with constraints beyond the lab. Danezis has said that when Libra/Diem did not progress as hoped, he left to pursue other ways of realizing high-performance blockchains. His former Meta colleagues Evan Cheng, Sam Blackshear, Adeniyi Abiodun and Kostas Chalkias joined him as founders of Mysten Labs in 2021.
Mysten’s founding team did not arrive as strangers assembled for a pitch. They had worked together on Libra at Meta. That history matters because the design of a blockchain is an exercise in coordinated disagreement. Engineers must decide what the system can do, what it will refuse to do, and which compromises are acceptable. At Mysten, Danezis serves as chief scientist. The title fits someone whose public explanations frequently return to architecture, limits and the experiment of putting an architecture before users.
A character, a hat and an auction
To explain Sui, Danezis reaches for a game character. Suppose the character owns a hat and wants to put it on. Both objects belong to the same person. The network can establish the action without making every participant agree on the order of a shared event. Now imagine an auction for hats. Several bidders can change the auction’s state, so the network needs consensus about what happened first. It is a child-sized example carrying an adult-sized systems design: a fast path for actions on owned objects, a consensus path for shared ones.
Sui Lutris, a protocol Danezis co-authored with colleagues, combines those paths. The aim is to avoid charging every action the cost of global agreement while preserving a common record where agreement really matters. On the surface, the choice is technical. In practice, it can decide whether a game action feels immediate or whether an application waits for the network to coordinate. Danezis’s example also makes clear that the developer must choose carefully. An object made shared when it need not be shared takes a slower route; an object treated as solely owned must obey the rules that make the fast route safe.
His answer to a question about Sui’s first months is particularly revealing. In a 2023 interview, he pointed to days with more than 60 million transactions and said most used the fast path. That was encouraging for a design whose use at scale had been uncertain. Then he described a problem. If a person, or the person’s software, sends two contradictory actions for the same owned object, the system can lock that object while it resolves the conflict. A phone and a computer acting at once could be enough. He spoke about the awkwardness plainly: developers worried about accidental locks, and the team was working on quicker recovery. A system had passed one test and revealed another.
The public conversation around blockchain often treats speed as a number on a scoreboard. Danezis’s account is less tidy and more useful. A fast path only matters if developers understand it and users can recover from mistakes. He described the practical consequence of a temporary lock as a real concern, not a footnote. That willingness to explain the unpleasant edge of a design is a rare service in a field where the temptation to speak only in projected throughput can be strong.
The next bottleneck
The work has continued beyond the original Sui design. Danezis co-authored the Mysticeti consensus research, which examined how to reduce the time needed for shared transactions. The paper reported a design that reached consensus in three message rounds and, in its evaluation, reduced latency after integration with Sui. Such numbers come from a defined research setting, but they identify the target: waiting. The person clicking a button may never hear the word “consensus”; they will notice the delay.
In 2025, Danezis appeared among the authors of a paper on Walrus, a decentralized storage network. Storage asks a different question from transactions. Data must remain available despite failures or uncooperative nodes, without requiring wasteful copies of everything everywhere. The Walrus paper describes RedStuff, an encoding approach meant to make recovery efficient while protecting availability. The subject is less immediately theatrical than a digital currency, yet it returns to the same stubborn promise: a network should do what it says even when individual machines do not.
Privacy remains part of the story. Danezis has argued that the right protections depend on the application. A game and a financial service do not expose the same information or carry the same risks. In his account, a blockchain can provide cryptographic building blocks, while application designers decide how to use them. He has pointed to zero-knowledge proof verification on Sui as one such building block. It is a measured answer from someone who spent years studying how generic privacy claims fail under specific scrutiny.
The path from Cambridge to Mysten Labs therefore reads less like a conversion than a change of venue. The audience grew from research peers to builders and ordinary users. The constraints became sharper. A design that seems elegant in a paper must survive impatient hands, conflicting devices, and demands no diagram can fully predict. Danezis keeps returning to that last test. His career is a reminder that the internet’s promises are made in language, written into protocols, and judged at the moment a person tries to use them.