Federated learning allows hospitals to train a shared clinical model without exchanging patient records, but its aggregation assumes that each participant is a distinct institution. In an open federation, an adversary can register many fake identities (a Sybil attack) and outweigh honest hospitals. Byzantine-robust aggregation mitigates such attacks only while the adversary is a minority. We ask whether binding participation to accountable institutional identity closes that gap, and at what cost. We present TBFL, a framework in which an accountable authority issues Verifiable Credentials to the Decentralized Identifiers of hospitals, and a smart-contract registry on a peer-to-peer ledger admits model updates only from credentialed senders. We evaluated a prototype on in-hospital mortality prediction with MIMIC-IV records split across simulated hospitals, comparing the identity gate with federated averaging, three robust aggregators, norm clipping, and a similarity-based defense. Robust aggregators held against a Sybil minority, and norm clipping only against random vectors; all failed when half the participants were Sybils trained on real records with flipped labels. The similarity-based defense stopped mutually similar Sybils but yielded to perturbed ones. The gate rejected all unauthorized submissions and maintained attack-free performance for every strategy up to 60% Sybils. It did not stop credentialed hospitals that flipped their own labels, whereas the gate followed by the coordinate-wise median withstood both. On a local node, on-chain checks added about 1% to training time; fees made training impractical on a public mainnet but affordable on a rollup. Identity gating complements robust aggregation rather than replacing it.
Figures & tables
Technologies
Evaluation
Study
Year
SSI
FL
Blockchain
Robust agg.
E1
E2
E3
E4
Houtan et al. [ 17 ]
2020
✓
✗
✓
✗
✗
✗
✗
✓
Cocco and Tonelli [ 18 ]
2024
✓
✗
✓
✗
✗
✗
✗
✓
López Martínez et al. [ 19 ]
2026
✓
✗
✓
✗
✓
✗
✗
✓
Kim et al. [ 11 ]
2020
✗
✓
✓
✗
✓
✗
✗
✗
Fung et al. [ 6 ]
2020
✗
✓
✗
✓
✗
✓
✓
✗
Table 1: Overview of related studies. ✓: present or supported; ✗: absent or not evaluated. E1: identity or ledger overhead measured or analyzed; E2: attacks evaluated experimentally; E3: comparison with Byzantine-robust aggregation; E4: healthcare setting.
Symbol
Meaning
Symbol
Meaning
K
number of honest clients (hospitals)
S
number of Sybil identities
wt
global model after round t
wkt
local model of client k in round t
nk
training samples of client k
Fk
local empirical loss of client k
μ
FedProx proximal weight
E,B
local epochs, batch size
I
trusted issuer
P
set of authorized addresses
ak
ledger address (DID) of client k
hkt
commitment SHA256(wkt)
Table 2: Notation.
Figure 1: System model of TBFL. Solid arrows are the honest protocol; dashed arrows are Sybil attempts, which the registry reverts and the aggregator discards. Only the digest hkt=SHA256(wkt) is written on-chain.
Figure 2: Admission protocol and one training round with the signed-credential registry. Blue numbers are the steps referred to in Section 5.1 ; orange letters are Sybil attempts (dashed arrows). Boxes on a lifeline are local actions. In the allowlist variant, the issuer itself calls authorizeWorker after verifying the credential off-chain.
R=100 (Exp. 1 and early stopping), R=50 (Exp. 2–5)
Table 6: Parameters of the experiments.
Model
AUC-ROC
AUPRC
Brier
Precision
Recall
TBFL (federated)
0.869±0.031
0.167±0.049
0.099±0.031
0.093±0.006
0.642±0.236
Centralized, same data
0.876±0.006
0.171±0.015
0.122±0.046
0.092±0.035
0.726±0.150
Local only a
0.852±0.008
0.144±0.006
–
0.090±0.003
0.578±0.040
Centralized, sensitivity to repeated patients b
Admission split
0.895±0.000
0.218±0.006
0.128±0.006
0.085±0.003
0.819±0.010
Patient split
0.895±0.003
0.221±0.011
0.136±0.004
0.080±0.004
0.838±0.013
Table 7: Predictive performance without attack (mean ± SD over five seeds; threshold 0.5). Upper block: global test set for the federation (prevalence 2.16%); baselines are trained for 30 epochs. A constant prediction of the prevalence has a Brier score of 0.021.
Figure 4: Learning curves of TBFL without attack on the global test set (mean ± SD over five seeds).
Figure 5: Final TBFL models after round 100 on the global test set (109,210 admissions, 2,360 deaths). In (b) and (c), each line is one seed and the dotted line is a random classifier.
Figure 6: Performance on each client’s local test set of the federated model and of a model trained only on that client’s data (seed 0).
k=1
k=2
Configuration
AUC-ROC
AUPRC
AUC-ROC
AUPRC
TBFL (gate + FedAvg)
0.746±0.231
0.103±0.078
0.411±0.346
0.031±0.032
TBFL + median
0.894±0.002
0.205±0.002
0.893±0.001
0.200±0.006
Table 8: AUC-ROC and AUPRC at round 50 with credentialed insiders (Experiment 5; mean ± SD over three seeds). From round 10, the k largest clients train on their own data with inverted labels; there are no Sybils. Attack-free references are in Fig. 7 .
Figure 7: AUC-ROC at round 50 under a Sybil attack starting at round 10, by defense and attack condition (mean ± SD over five seeds unless marked). S Sybil identities join 10 honest clients; robust aggregators receive the true S , capped as in Table 6 . The color scale is centered at 0.8: orange cells indicate defenses that failed, while blue cells indicate defenses that maintained their attack-free accuracy. –: condition not run. ∗ Three seeds. † The gate rejects Sybils by identity before their update content is observed, so the adaptive attack yields the same admitted set as the identical attack.
Adaptive, S=10
Label flipping
Defense
σ=0.1
σ=0.3
S=5 (33%)
S=10 (50%)
FedAvg (unprotected)
–
–
0.741±0.156
0.250±0.024
Coordinate-wise median
0.500±0.000
0.500±0.000
0.883±0.005
0.186±0.039
Trimmed mean
–
–
0.879±0.005
0.191±0.046
Krum
0.893±0.003
0.896±0.003
0.889±0.003
0.113±0.014
Norm clipping
–
–
0.737±0.139
0.202±0.031
Table 9: AUC-ROC at round 50 under further Sybil strategies (mean ± SD over three seeds). Left: adaptive Sybils with smaller noise, S=10 ( σ=0 and σ=1 are in Fig. 7 ). Right: label-flipping Sybils (Experiment 4).
Figure 8: Global AUC-ROC per round under identical Sybils joining at round 10 (dotted line). Mean ± SD over five seeds.
Figure 9: Content-based defenses against identity-based gating. (a) AUC-ROC at round 50 against the share of identical Sybils (legend at top). The dotted line marks 50%. (b) Aggregation weight that FoolsGold assigns to adaptive and identical Sybils. (c) AUC-ROC at round 50 for each content-based defense under identical and adaptive Sybils. (d) AUC-ROC at round 50 under label-flipping Sybils (three seeds). Bars: mean over seeds (five in (c)).
Attempt
Goal
Registry
Accepted
Self-authorization by a non-issuer
G1
both allowlists
0 / 2
Commitment from an unauthorized address
G1
all three variants
0 / 60
Credential signed by a key other than I
G2
signed credential
0 / 1
Valid credential of another subject (replay)
G2
signed credential
0 / 1
Expired credential
G2
signed credential
0 / 1
Commitment after revocation
G2
signed credential
0 / 1
Table 10: Attack attempts against the registry and their outcomes (accepted/attempted). Upper block: registry benchmark with ten clients over 20 rounds; lower block: all 41 TBFL runs with Sybils in Experiments 2 to 4, of which 29 also ran the G4 check; the 12 earliest runs of Experiment 2 predate it.
Network
Registry
Contract address
Ethereum Sepolia
Allowlist ( FLRegistry )
0x54e698E73F26e18B17071BA471Cc6a11B6d5D57f
Ethereum Sepolia
Signed credential ( FLRegistryZK )
0x2730b56bc97560D342a2032F6C4239c56eB5667F
Arbitrum Sepolia
Allowlist ( FLRegistry )
0xc4Ed367EEA1a699435F8351039b21e154b728476
Arbitrum Sepolia
Signed credential ( FLRegistryZK )
0x759F85E519Cb6ABcFCe14A4833ef8C071490857E
Table 11: Registries deployed on public test networks in October 2026. Transactions can be inspected at https://sepolia.etherscan.io and https://sepolia.arbiscan.io .
Local node
Arbitrum Sepolia
Ethereum Sepolia
Gas schedule
pre-Glamsterdam
Arbitrum
Glamsterdam
Deployment gas, A / S
723,368 / 1,935,532
723,368 / 1,935,532
4,964,532 / 13,611,116
Authorization gas, A / S
46,333 / 80,134
46,333 / 80,134
128,353 / 250,062
Commitment gas, A / S
46,873 / 48,869
46,873 / 48,869
62,273 / 64,269
One-transaction confirmation
0.05 s
1.2 s (1.0–1.6)
10.6 s (6.8–22.5)
Five-client round, A / S
–
9.7±0.2 / 9.6±0.3 s
15.7±7.7 / 17.5±10.7 s
Table 12: Registries deployed on public test networks (five clients, ten rounds per registry, October 2026), compared with the local node. Gas values are medians; A/S denotes the allowlist and signed-credential registries. Contract addresses are listed in Table 11 .
Allowlist
+ events
Signed VC
Deployment
723,368
759,286
1,935,532
Authorization, once per client
46,333
47,494
80,134
Commitment, per client and round
46,873
47,256
48,869
Revocation
–
–
52,724
Commitments per round, 10 clients
468,730
472,560
488,690
L1, 20 gwei: per round / 100 rounds
28.12/2,812
28.35/2,835
29.32/2,932
Table 13: Measured gas (upper block), derived cost in US dollars and time overhead. Gas values are medians from the registry benchmark with ten clients; per-round costs use the median commitment and exclude one-off deployment and authorization. USD values assume ETH = US$3,000; the rollup value excludes layer-1 data fees.
Figure 10: Efficiency of on-chain admission. (a) Median gas per operation for the three registry variants. (b) Cumulative cost of the allowlist over 100 rounds on Ethereum L1 and on Arbitrum (log scale). (c) Local training time versus commitment latency on the local node (log scale).
Property
PKI and mutual TLS
Ledger registry (TBFL)
Excludes uncredentialed identities
yes, under (A1)
yes, under (A1)
Enforced by
aggregator operator
contract, run by every node
Record of admissions
operator logs
tamper-evident, public to members
Revocation visible to
relying party (CRL, OCSP)
every member (contract state)
Several aggregators
separate trust stores
shared admission state
Update–sender binding
needs a separate signed log
commitments on the ledger
Table 14: Identity gating with a conventional PKI at the aggregator and with the ledger registry of TBFL. Both rely on the same accountable issuer (A1).
Dept. of Applied Mathematics and Theoretical Physics University of Cambridge Cambridge, UK · Translational AI Laboratory, Dept. of Laboratory Medicine Amsterdam UMC Amsterdam, The Netherlands · Precision Health University Research Institute Queen Mary Univ. of London London, UK +3
Department of AIML Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology Chennai, India · Assistant Professor, Department of AIML Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology Chennai, India · Professor, Department of AIML Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology Chennai, India