Proof of personhood

The approach originally proposed by Borge et al. was to use in-person pseudonym parties as a basis to create anonymous one-per-person tokens periodically without requiring any form of identity verification.[3][5] The encointer project adapts this approach by asking participants to meet in small groups simultaneously at randomly-chosen places, to verify each other's physical presence.[7]

One drawback of this approach is the inconvenience to participants of going to designated physical locations at specific times, especially for participants with conflicting responsibilities at those times. Another issue is the challenge of organizing federated pseudonym parties in multiple locations simultaneously while allowing each group to verify that all other groups are organized honestly without inflating the number of digital credentials they issue.

Another approach, related to the PGP Web of Trust, relies on users forming a social network to verify and attest to each other's identities.[8] UniqueID incorporates biometric verification into the social network approach.[9]

One criticism of the social network approach is that there is no straightforward way for a participant to verify that a social connection has not created other Sybil identities connected to and verified by other, disjoint sets of social contacts. A related challenge is that Sybil detection based on graph analysis make certain assumptions about the behavior of a Sybil attacker, and it is not clear that real-world social networks satisfy these assumptions.[10] Finally, graph-based Sybil detection algorithms tend to be able to detect only large, densely-clustered groups of Sybil nodes in a social network, leaving small-scale attacks difficult or impossible to distinguish by graph structure alone from legitimate users' connectivity structures.

Another approach requires participants to have verified identities, but to hide or anonymize those identities in subsequent use. One criticism of this approach is the privacy and surveillance risks inherent in such databases, especially biometric databases, and the level of trust users must place in the verification service for both Sybil protection and privacy of their identity information. Other critics highlight that facial recognition systems fail on a global scale due to insufficient facial entropy.

Apple, who are known for implementing a facial recognition feature into the iPhone, attempts to protect users' privacy with the Secure Enclave. The mathematical structure of a user's face captured by the TrueDepth camera does not leave the user's device, increasing the privacy and protection of personal information.[11][12] However, some concerns have been raised in regards to the level of security of the facial recognision on the devices. For example, there have been cases where family members were mistakenly recognized as their siblings.[13]

Even with decentralized privacy protections, a criticism of this approach is the inconvenience and cost to users of verifying strong identities, and the risk of potential exclusion of users who do not readily have or cannot afford the requisite identity documents, are reluctant to participate due to privacy and surveillance concerns, or are wrongly excluded by errors in biometric tests.[14]

To resolve security concerns over using biometrics for proof of human existence, encrypting the biometrics data through cryptographic models isn't enough. One technique proposed to solve this is to use homomorphic encryption along with zero-knowledge proof to encrypt biometrics data in a way that original biometrics data never leaves the device of the user. Instead, the decentralized network is provided only with the relevant information to verify if a person is a real human being through liveness detection and is registered on a network.[15]

Another proposed class of approach extends the CAPTCHA principle of using Turing tests to the unique human verification problem. The Idena network, for example, assigns participants to verify each other using flip tests.[16] Criticisms of this approach include the inconvenience to users of solving Turing tests, and whether artificial intelligence and deepfake technologies will soon be able to solve such tests automatically or convince real participants that a synthetic user is human during a verification interaction.