Trusta’s “Proof of Innocence Program”— Fighting Back Against Poisoning

Abstract

• The recent Connext Sybil Hunter Program has caused discontent and heated discussions in the community, with individual community members threatening to poison Connext airdrop addresses, especially those in the top 10% of transactions on zkSync. The reality is that some users said they have been poisoned. Trusta Labs, an on-chain data analytics and security platform, discusses the two main tactics of sybil poisoning attacks, forced Association/clustering and label propagation tactics, and analyzes in detail a real-life Connext sybil poisoning case that occurred on Polygon.
• This type of poisoning may cause many innocent addresses to be incorrectly labeled as Sybils, which can hurt users and the entire community, as well as the trust between the community and the project. Therefore, Trusta Labs has launched the “ Proof of Innocence Program” (PoIP) with the community and the project to fight against poisoning and let the poisoned addresses prove their innocence. In the PoIP , users submit relevant information to be verified by Trusta with a combination of manual review and AI analysis, and users get feedback within 1 day. This data will be shared with the project side and the anti-sybil team to prevent sybils from identifying false positives and better protect users.

Connext Sybil Hunter Program & Poisoning

Connext is a crosschain liquidity network that enables fast, fully-noncustodial transfers between EVM-compatible chains and L2 systems. Let’s outline the timeline of Connext airdrop and sybil hunting:

1. On August 18th, Connext announced a cross-chain airdrop of its xERC20 $NEXT token.
2. On August 24th, Connext launched the Community Sybil Hunter program, like HOP and SAFE have done.
3. From August 24th to September 1st, community members identified and reported Sybil attackers.
4. As of September 1st, the program collected ~600 reports involving ~20K addresses out of 62,070 candidates (35%).

While the intention of selecting and rewarding valuable real users is good, we have seen a lot of controversy in the community about the sybil hunter program. Among them, “poisoning” has become a hot topic, there are some reported sybil address users threatened to poison other wallet addresses, their purpose is to mess up the whole sybil reporting work and airdrop program. The reality is that some users have said that they were indeed poisoned.

Poisoning Tactic Analysis

Based on on-chain data analysis and security risk control experience, Trusta Labs analyzed and discussed the two main techniques and strategies of sybil poisoning attacks: Forced Association/Clustering and Sybil Propogation, with real poisoning cases.

Poisoning Tactic I (Forced Association/Clustering): The poisoner addresses use tools like batch operation scripts or disperse.app to conduct mass token transfers. They make many small transfers to a bunch of innocent addresses within a short period of time. All of the forged transfers involve sending the same token in tiny amounts.

Through the poisoners’ mass token transfers, all of the unrelated addresses are forcibly associated together into a cluster. This forced clustering is based solely on the addresses sharing a single poisoner address source, even though the addresses are actually unconnected to each other.

Poisoning Tactic II (Sybil Propogation): Label propagation is a graph mining algorithm based on the intuition that nodes closely connected in a graph tend to have the same label.

In the diagram, addresses in the square are already labelled as Sybils, such as Chain-Like Sybil. A bad actor can use these Sybil addresses as poisoners to intentionally make transfers to innocent addresses to propagate the Sybil label to them.

Sybil Propagation relies on existing Sybil addresses to spread their label to other addresses through transfers. This limits which actors can deploy this poisoning tactic. In contrast, Forced Association does not require pre-existing Sybil addresses. Any address can be used as a poisoner to artificially manufacture false Sybil activity patterns. This makes it much easier and lower cost to execute.

In the process of analyzing the data on the chain, Trusta found a case of Connext poisoning on Polygon, and through a detailed analysis of this case, explained the above poisoning techniques.

A Real Case of Connext Poisoning

Trusta discovered a Connext poisoning case effortlessly. As shown, Poisoner 0x6ab used disperse.app to batch transfer to seven innocent addresses. We deem this a poisoning case for the following on-chain reasons:

1. The Poisoner’s Polygon scan shows after being deposited ~1 Matic from OKX, its only action was batch transfers to these 7 addresses.
2. The poisoning transfers occurred from 2023–08–25 05:49:40 to 2023–08–25 05:52:12, during the Sybil hunting program.
3. The Poisoner conducted 7 rounds of transfers, with each round sending 0.0001 MATIC to each of the 7 addresses. Since the Poisoner could make multiple transfers to the same address in a given round, there were a total of 180 transfers across all 7 rounds. The full list of 180 poisoning transfers can be found in the provided google link..
4. All 7 addresses are Connext airdrop candidates. Sybil Report #589 accused them of being a Sybil cluster using these transfers as evidence.
5. There are no direct transfers detected between any of these 7 addresses. They have not transacted with each other outside of the poisoning attacks.
6. We analyzed the addresses’ Polygon activity statistics. As the table shows, the seven differ completely in factors like first deposit source, transaction dates, contracts interacted with, gas fees, and active weeks/months. This extreme variance means they cannot belong to one Sybil cluster operated by one entity.

Through this analysis, we conclude this is an actual case of poisoning related to the Connext airdrop, employing the tactic of forced association/clustering.

The Proof of Innocence Program (PoIP) by Trusta

Trusta has noticed many unrelated addresses wrongly labeled as Sybils, not just in the Connext airdrop but also indiscriminate attacks on innocent wallets. Poisoning is misbehavior that destroys trust. While Trusta’s solution has enabled projects to reward valuable users, we must also ensure fair treatment for users.

To unite against poisoning, Trusta has launched the “Proof of Innocence Program (PoIP)”. It lets poisoned addresses prove their innocence. If your address was infected:

1. Go to the PoIP entrance and provide poisoning details like your address, poisoner, transaction hash, and chain. (https://medium.com/r/?url=https%3A%2F%2Fdocs.google.com%2Fforms%2Fd%2Fe%2F1FAIpQLSe_1dl6ocyhnDWUtm9BBvmWDGL_rDjhc9NNpfHXff2XhXL5eg%2Fviewform)
2. Trusta manually and algorithmically judges if it is poisoned.
3. You will receive the decision within a day via email.
4. The data will form a database proving your address is unrelated to poisoners! TrustScan will use it to prevent false positives by ignoring poisoned relationships. Trusta will make the data open to all project parties and anti-sybil teams to prevent addresses being wrongly recognized as Sybils only due to poisoning transfers.

Both TrustScan (the real user identification service, which is used by several headline communities such as Gitcoin Passport) and Trustgo (the on-chain value scoring product https://trustgo.trustalabs.ai/), will incorporate this data to continually improve the accuracy of user identification and segmentation results. At the same time Trusta will also make this data available to all project parties and anti-sybil teams to prevent these innocent addresses from being incorrectly identified as sybil addresses due to poisoning attacks.

More Trust, Less Friction!

Trusta Offical Website: https://www.trustalabs.ai/

Titter Handle: @trustalabs

Discord: http://discord.gg/t64shY5DwT