On September 20, security teams and project developers detected anomalous minting transactions on the Ethereum network, revealing that the same hacker who had previously attacked Fetch.ai and NuNet had once again targeted SingularityNET's cross-chain bridge using the same technical approach. The root vulnerability was not in the frontend contract logic, but in the theft of a backend authorization signing key used after cross-chain bridging: after obtaining this "master key," the attacker could initiate formally valid but unauthorized on-chain minting requests. According to an on-chain analysis report, the hacker unauthorizedly minted approximately 260 million AGIX and around 53.83 million WMTx/WMTX on Ethereum, subsequently transferring some of these assets to a wallet address under their control.
Key attack figures: 260 million AGIX + 53.83 million WMTx unauthorizedly minted; hacker controls ~$16.77 million in crypto assets (~198.3 million AGIX worth ~$14.42 million); three consecutive attacks on Fetch.ai, NuNet and SingularityNET; compromise of backend signing keys for cross-chain bridges.
According to disclosures from security agencies such as PeckShield, the address behind the SingularityNET cross-chain bridge incident closely matches addresses used in the previous attacks on Fetch.ai and NuNet, demonstrating that the same entity launched attacks within a short time frame to carry out a "sequence of three attacks" around the same ecosystem narrative. The first two attacks demonstrated that this attacker not only understands the contract structures and deployment practices of the respective projects, but also comprehends the architecture of the underlying infrastructure, allowing them to reuse their technical stack and criminal pathways across different protocols.
As the attack target expanded from Fetch.ai and NuNet to the SingularityNET cross-chain/bridge contract exposed on September 20, the event became more than just an incident affecting a single project; it began to resemble a concentrated stress test of the entire technology stack and security governance model associated with AI protocols. All three attacks were rooted in similar contract designs and backend authorization mechanisms, indicating that as long as vulnerabilities remain in core key management or bridging logic, a single attack path can be reused across the broader ecosystem, shifting the risk from one token to a wider group of interconnected protocols.
Public information still cannot confirm the attacker's real identity or organizational affiliation. However, the attack pattern demonstrates a high level of technical expertise and a deep understanding of cross-chain bridge architecture. The attacker demonstrated the ability to identify and exploit the same vulnerabilities across different projects, suggesting either insider knowledge or careful reverse engineering of publicly available infrastructure. The coordination of the attacks within a narrow time window also points to strategic planning rather than opportunistic actions.
In a traditional cross-chain bridge design, the backend authorization signing key acts as the "final arbiter": when an asset on one chain is locked or burned, the backend service generates a signed message and submits it to the bridge or token contract on the destination chain to trigger the minting or unlocking of the corresponding asset. The on-chain contract itself does not know what happened in the real world; it only verifies the validity of the signature. This key therefore effectively serves as a single point of trust for the entire cross-chain and minting process. Once compromised, every contract relying on it is immediately exposed to the same attack path.
In this incident, the theft of the SingularityNET cross-chain bridge's backend authorization signing key became the critical turning point. After gaining control of the key, the attacker could initiate outwardly "legitimate" minting requests to the AGIX and WMTx/WMTX contracts on Ethereum, generating signatures indistinguishable from those produced during normal cross-chain operations. The on-chain contract could not distinguish whether the request represented a user-initiated completion of a cross-chain transaction or malicious arbitrary minting.
As a result, approximately 260 million AGIX and around 53.83 million WMTx/WMTX were unauthorizedly entered into the ledger, disrupting the originally designed total token supply and issuance schedule, diluting the expected supply of all existing token holders and shaking the entire ecosystem's confidence in the cross-chain bridge and its underlying signature management system. Given that the path and timing of the key leak have not been publicly disclosed, this trust gap is unlikely to be automatically resolved in the short term. The compromise of a single key created a cascading effect that affected multiple related projects and undermined confidence in the security of the entire ASI Alliance infrastructure.
The scale of the unauthorized minting began to emerge, moving from "expected dilution" to numerical reality. According to AiCoin's analysis, Fetch.ai publicly released an on-chain analysis on the ASI:One platform indicating that the attacker unauthorizedly minted approximately 260 million AGIX and around 53.83 million WMTx/WMTX on Ethereum using the stolen cross-chain bridge signing key. Cumulatively, this incident created potential selling pressure of more than 313 million related tokens on paper, rewriting the total supply curve originally built into the protocol into a different version.
Even more striking is the attacker's current asset profile. The same report states that the attacker's address controls around $16.77 million in crypto assets, including approximately 198.3 million AGIX worth about $14.42 million, with the remaining value distributed among other assets. This indicates that a significant proportion of the "new coins" minted had already been transferred to a wallet controlled by the attacker, which on-chain analysis flagged as a potential cash-out route.
However, it should be emphasized that all of the figures above come from this single report, which has not yet been finalized, and Fetch.ai noted in the text that further adjustments and modifications may be made based on new on-chain evidence. At present, public materials do not provide changes in the price or total market capitalization of AGIX and WMTx/WMTX before and after the incident, nor do they explain whether the project teams will deal with this batch of unauthorized tokens through burning, rollback or other technical and governance measures. In the absence of these key parameters, outside observers can only treat this preliminary analysis as a foundational reference point outlining the losses, without drawing easy conclusions about market impact or long-term spillover effects.
Following the incident, Fetch.ai launched its first on-chain analysis report on the ASI:One platform, focusing on tracing the complete path from the theft of the authorization key to assets flowing into cash-out wallets. The report identified key on-chain steps in chronological and transactional order: how the backend key was misused to initiate outwardly valid minting requests, how the cross-chain bridge contract unauthorizedly minted approximately 260 million AGIX and around 53.83 million WMTx/WMTX, and how these newly minted tokens were transferred in batches to wallets controlled by the attacker.
At the same time, the report clearly stated that it was "not the final version," indicating that additional transaction samples and address profiles would be added later, while for now it serves primarily to provide the community and related projects with a verifiable baseline timeline of events rather than to declare the investigation complete. While conducting tracing and gathering evidence, Fetch.ai and SingularityNET jointly pressed the "emergency stop button," disabling confirmed affected wallets and related contracts, with the primary goal of preventing the attacker from continuing to invoke the cross-chain bridge or transfer tokens still under their control, thereby preventing artificially minted tokens from spreading to a wider range of on-chain participants.
According to AiCoin data, the public can currently see only a directional description of this containment action: the full list of wallets and contracts has not been disclosed, and the specific technical measures used for deactivation and the actual effects of the blocking have not been transparently revealed. In other words, the ASI Alliance has moved from responding to "individual attacks" toward a coordinated response, but the overall focus remains on stopping the bleeding and clarifying the attack path; until more on-chain data, technical assessments and plans for subsequent handling are revealed, the true risk profile of this arbitrary minting incident has not yet been fully defined.
From the initial cross-chain security alarm to subsequent observations — from the stolen authorization signing key to unauthorized minting on Ethereum, and then to the same attacker continuously targeting Fetch.ai, NuNet and SingularityNET — this incident exposed the risk of a cross-chain bridge backend key as a single point of failure in exceptionally direct terms: compromising one key was enough to arbitrarily generate hundreds of millions of AGIX and WMTx/WMTX on-chain, undermining the trust foundation of the entire tokenomics model.
For the high-profile AI narrative ecosystem, an even more concerning signal is that while the industry eagerly invests in models, computing power and alliance narratives, it has yet to establish a consistently high security standard for cross-chain architecture and key governance. Going forward, several points deserve close attention: first, whether Fetch.ai can further reconstruct the complete path from the key theft to the attacker's approximately $16.77 million in assets through continuous updates to its on-chain analysis; second, how projects such as SingularityNET will rebuild their cross-chain bridge architecture and key management to genuinely reduce backend single-point control; and third, whether the project teams will disclose their plans for handling the unauthorized tokens, because how they deal with 260 million AGIX and 53.83 million WMTx/WMTX will directly affect market assessments of their governance capabilities and risk controls over the long term.
At the broader industry level, this attack has almost forced the entire cross-chain infrastructure sector to re-examine whether institutional upgrades such as multi-signatures, hardware security modules and permission layering are actually in place, and whether transparent and verifiable on-chain responses can be provided during similar incidents. The trajectory of these variables will determine whether this arbitrary minting crisis is ultimately remembered as a lesson or as the beginning of a prolonged trust deficit.
“Security is not a product, but a process. In the world of cross-chain infrastructure, the compromise of a single key can destroy trust in an entire ecosystem, making key management the most critical component of security.”
— Andreas Antonopoulos, blockchain security expert
