Security, audits, incidents and dependency risks

A dated research snapshot

We checked the sources for this chapter on 17 Sept 2026. Use the linked sources for current availability, terms and network details. See subsequent updates

Arc has a public security entry point and a HackerOne bounty program, with rewards advertised up to $1 million. Only the currently authorized targets and testing methods qualify. The original open-source/bounty announcement had a defined scope; permission to attack mainnet or third-party applications should never be inferred from the existence of a bounty. S020 S021 S022

The node and consensus repositories are useful because they expose implementation, issue tracking, licensing and security caveats. Their README language includes alpha/development or audit-status wording that may not be synchronized with the mainnet announcement. That discrepancy is a diligence item. It is not sufficient evidence either that no audits occurred or that all production code has completed independent audits. S017 S018

I did not establish a complete public set of final mainnet audit reports tied to exact production commits, all deployed privileged roles, or a fully reconciled incident archive. A reliable security assessment would need those artifacts, not just auditor logos or grant-funded audits at the application layer.

Principal risk domains

Consensus and implementation: safety/liveness under fault assumptions, release quality, validator software diversity, operational incidents and recovery procedures.

Stablecoin and custody dependencies: issuer controls, fiat redemption, custody of represented assets and the consequences of restrictions on the gas asset.

Application and market risk: contract upgrades, lending parameters, price feeds, thin liquidity, liquidation execution, leverage and tokenized-asset permissions.

Crosschain risk: attestation/verification, route configuration, address mistakes, destination availability and recovery paths.

Privacy/agent risk: enclave trust, key management, access policies, software-agent authorization, credential theft and unintended spending. AI-generated code or an “agentic” product label does not constitute an audit.

These are analytical risk categories grounded in the published architecture and service model, not allegations that a specific undisclosed incident occurred. S009 S014 S075 S088 S090

Sources & references

Documents cited in this chapter.

  1. Arc securityarc.io
  2. Arc HackerOne programhackerone.com
  3. Open sourcing Arc and bug bountyarc.io
  4. Arc node repositorygithub.com
  5. Malachite repositorygithub.com
  6. System overviewdocs.arc.io
  7. Opt-in privacy designdocs.arc.io
  8. Arc Portal termsarc.io
  9. USDC legal termscircle.com
  10. Circle developer documentationdevelopers.circle.com
  11. Original sourcegist.github.com