Traditional privacy frameworks treat personal data like a private commodity. The underlying assumption is simple: your data belongs to you, and if you choose to share it, you are the only one who bears the consequences.
However, a growing body of research demonstrates that this individualistic model is fundamentally flawed. In reality, data is highly correlated across social, behavioral, and biological networks. When you disclose your information, you inadvertently leak sensitive data about others who never gave their consent. This phenomenon is known as interdependent privacy.
To understand why interdependent privacy occurs, how it causes systemic market failures, and how researchers are working to mitigate it, we must look at the mathematical and structural mechanisms that link our data together.
The Mathematical Engine of Inference
Why does sharing one person's data expose another? The answer lies in statistical dependencies. If two data points are correlated, observing one reduces the uncertainty of the other.
This mechanism is most potent in genomics. Your DNA is not an isolated string of code; it is a mosaic of genetic material shared with your biological relatives. As Jun Sakuma explains in Genome and Privacy, an individual's genetic privacy is intrinsically linked to the data-sharing decisions of their kin.
An adversary can exploit these biological links using two primary mechanisms:
- Mendelian Inheritance: The laws of genetics dictate exactly how alleles are passed down from parents to offspring. If an attacker knows the genetic profiles of your parents, your own genetic profile is heavily constrained.
- Linkage Disequilibrium: Even within a single individual's genome, genetic markers (Single Nucleotide Polymorphisms, or SNPs) are statistically correlated based on their physical proximity on a chromosome. As noted by Erman Ayday et al. in Towards Quantifying and Preventing the Leakage of Genomic Data Using Privacy-Enhancing Technologies, revealing a single genetic variant can inadvertently expose unobserved, sensitive traits elsewhere in the DNA sequence.
[ Father's DNA ] [ Mother's DNA ]
\ /
\ /
▼ ▼
[ Your Hidden DNA ] ◄─── (Inferred via Mendel's Laws)
To operationalize these biological constraints, attackers use graphical models, such as Bayesian networks. By treating family members as nodes and inheritance laws as probabilistic edges, adversaries can run belief propagation algorithms. As highlighted by Kenneth W. Goodman et al. in Between Access and Privacy: Challenges in Sharing Health Data, these algorithms simulate how information flows through a family tree, allowing an attacker to reconstruct the private health status or genetic profile of a non-consenting target.
This vulnerability is magnified by the sheer scale of modern databases. Ardalan Naseri et al. demonstrated in Personalized genealogical history of UK individuals inferred from biobank-scale IBD segments that by analyzing Identity-By-Descent (IBD) segments across large biobanks, a significant portion of an individual's genome can be imputed using data from distant relatives, even if the target individual has never provided a sample.
The Non-Linear Collapse of Privacy
A common misconception is that privacy loss scales linearly with the amount of data leaked. In interdependent systems, privacy behaves like a depleting resource that experiences a sudden, non-linear collapse.
In his thesis, When Others Impinge upon Your Privacy - Interdependent Risks and Protection in a Connected World, Mathias Humbert showed that because of the high correlation between relatives' DNA, disclosing just a small fraction of one member's genetic information can compromise more than half of the collective privacy for the entire family group.
This compounding effect was quantified by Didier Dupertuis et al. in KGP Meter: Communicating Kin Genomic Privacy to the Masses. While a single distant relative's data might cause minimal friction, the combined data of a few close relatives (such as parents or children) can instantly reduce an individual's genomic uncertainty by over 80%.
This means that a tiny fraction of a population can compromise the anonymity of everyone else. As Jia Guo et al. point out in A game theoretic approach to balance privacy risks and familial benefits, once a small threshold of participants joins a genomic database, nearly every individual in that population can be identified through familial links, rendering individual non-participation ineffective.
The Economic and Strategic Tragedy
When privacy is interdependent, individual decision-making breaks down, creating severe economic and behavioral distortions. Economists and computer scientists model these interactions using game theory.
In Too Much Data: Prices and Inefficiencies in Data Markets, Daron Acemoğlu et al. explain that data sharing generates massive negative externalities. Because a platform can infer your information from your peers, your own data becomes less valuable to the buyer. This mechanism depresses the market price of data, leading to a market failure where individuals share too much data for too little compensation.
[ User A Shares Data ] ───► [ Platform ] ◄─── [ User B Withholds Data ]
│
▼
(Platform infers User B's
data anyway, depressing
the value of privacy)
This dynamic creates strategic dilemmas within families. In Interdependent Privacy Games: The Case of Genomics, Erman Ayday et al. show that the benefits of data disclosure (such as contributing to medical research or discovering ancestry) are captured entirely by the individual, while the privacy costs are distributed across non-consenting relatives. This imbalance means that even purely altruistic intentions can lead to inefficient outcomes that compromise the collective security of the entire family network.
Beyond Biology: Interdependent Privacy in AI
While genomics provides the clearest physical proof of interdependent privacy, the same mathematical vulnerabilities apply to digital and social networks.
When users interact with modern artificial intelligence systems, they often upload datasets containing information about their peers. Tianshi Li 0001 et al. explored this in Shaping the Emerging Norms of Using Large Language Models in Social Computing Research, noting that interactions with AI-enabled services frequently lead to the accidental disclosure of sensitive information regarding third parties. Traditional informed consent protocols are fundamentally unequipped to handle these multi-party data spillovers, as they assume a strict boundary between individual and collective data.
Engineering Solutions for Collective Privacy
Because simple data redaction fails to protect against correlation-based inference, researchers are developing specialized privacy-enhancing technologies (PETs) to manage interdependent risks.
One approach involves treating data disclosure as a multi-party negotiation. In Reconciling Utility with Privacy in Genomics, Jean-Pierre Hubaux et al. introduced a Genomic-Privacy Preserving Mechanism (GPPM). This framework uses combinatorial optimization to maximize the utility of shared data while strictly respecting the privacy "red lines" of an individual's entire family tree.
Other frameworks focus on sanitizing data before it is queried or published:
- Cryptographic Safeguards: The authors of Practical Solutions for Protecting Individual Genomic Privacy demonstrated that combining homomorphic encryption with graphical models allows medical centers to run diagnostic queries on encrypted DNA without ever exposing raw genetic sequences.
- Coordinated Defense: Rajagopal Venkatesaramani et al. showed in Enabling tradeoffs in privacy and utility in genomic data Beacons and summary statistics that defenses must account for correlations rather than treating markers independently. Their framework pairs variant masking with statistical noise to block adaptive attackers who exploit neighboring visible data points to reconstruct hidden information.
Ultimately, these solutions require a paradigm shift. As synthetic data generation and large-scale biobanks continue to expand, computer scientists must move away from individualistic privacy models. As Bristena Oprisanu et al. warn in On Utility and Privacy in Synthetic Genomic Data, failing to account for familial and network correlations leaves non-participating individuals highly vulnerable to indirect data leakage. Protecting privacy in a connected world requires architectures that recognize data for what it truly is: a shared asset.
Key Foundational Work
Towards Quantifying and Preventing the Leakage of Genomic Data Using Privacy-Enhancing Technologies (2013) — contributed 3 concept elements to this direction.
The Chills and Thrills of Whole Genome Sequencing (2013) — contributed 2 concept elements to this direction.
To Release or Not to Release: Evaluating Information Leaks in Aggregate Human-Genome Data (2011) — contributed 1 concept elements to this direction.