io.net Disclosed 1.8 Million Fake GPUs in Network-Capacity Reckoning

On April 28, 2024, io.net founder and chief executive Ahmad Shadid said the decentralized GPU network had expelled Sybil attackers after approximately 1.8 million fake GPUs attempted to connect over the preceding 10 days. The disclosure followed questions about how many working processors the service could actually supply and a separate April 25 breach that let malicious users alter device metadata displayed by its explorer.

Shadid said the team had spent 120 hours removing spoofed devices, deploying security patches and introducing a new security model. He described the infrastructure as fully operational, while warning that visible GPU supply could remain temporarily reduced as legitimate suppliers updated their software and rejoined.

The April 28 statement mattered because io.net was selling an institutional promise: that token incentives and verification could turn independently owned graphics processors into dependable cloud infrastructure. The episode showed that a DePIN network's headline capacity could be distorted when its rewards system counted identities or heartbeats that did not represent usable hardware.

Two related failures, not one confirmed compromise

The company's records describe two connected but distinct problems. First was the effort to manufacture counterfeit workers and collect prospective rewards. Shadid characterized that campaign as a Sybil attack and put the attempted fake-device spike at about 1.8 million. That figure was supplied by io.net; no independent forensic record reviewed for this reconstruction validates it.

Second was the April 25 metadata incident. In an April 27 technical report, io.net said an explorer API had exposed user identifiers associated with device identifiers. A different worker API then accepted an owner's user identifier in a header without user-level authentication, enabling authenticated attackers to change device names and displayed up-or-down status.

The company said monitoring detected an unusually high rate of writes to the GPU metadata API at 1:05 a.m. Pacific time on April 25. Its on-call staff contacted developers at 1:16 a.m., and remediation began within 10 minutes of detection. The team temporarily disabled metadata updates, accelerated a move from a universal authorization token to user-specific authentication, and reported completing the data migration in six hours.

Those are company-reported response times, not independently audited measurements. io.net also said the attackers could not access the underlying GPU hardware, run compute jobs on it or obtain user or device data through the metadata attack. The public record available on April 28 did not include an outside security assessment capable of proving those negative claims.

Why the capacity dispute carried weight

IO Research, the company behind io.net, had announced a $30 million Series A funding round on March 5, 2024. Its pitch was to aggregate underused GPUs into clusters for artificial-intelligence and machine-learning workloads. That made the difference between registered, verified and deployable processors central rather than cosmetic: customers needed real compute, while suppliers expected rewards to distinguish genuine capacity from fabricated workers.

The incident also exposed a basic incentive-design risk. When a future token distribution rewards apparent participation before identity and hardware checks are mature, attackers have a reason to simulate scale. In that setting, a large node count can measure the aggressiveness of reward farming as much as the service's productive capacity.

What April 28 did not establish

The April 28 postmortem established that io.net acknowledged the spoofing campaign, said it had applied countermeasures and accepted that its earlier presentation of network capacity had caused confusion. It did not establish how many GPUs were independently verified and ready to accept customer jobs at a single consistent measurement time. Nor did it quantify rewards wrongly accrued, customer workloads disrupted or financial losses.

No token-price or market-return claim is made because the reviewed contemporaneous record does not provide a traded instrument and clean event window for the disclosure. The defensible conclusion is narrower: io.net's response turned hardware verification and capacity transparency into immediate credibility tests for token-incentivized compute networks.