Safe Wallet Theft: Lessons from the $6.45M Loss10 min read2,091 words

Safe Wallet Theft: Lessons from the $6.45M Loss

A Safe wallet reportedly lost $6.45 million after a likely key compromise. See how permissions shaped the incident and what wallet teams can review.

Safe wallet theftcrypto wallet securityZodiac Roles Modifiermultisig permissionsprivate key compromise
Safe Wallet Theft: Lessons from the $6.45M Loss

The $6.45M Safe Wallet Loss Was a Permissions Problem, Too

A crypto wallet can run smart contracts correctly and still lose millions. That's the key distinction behind the reported $6.45 million loss from a Safe wallet. Zodiac's co-founder said the transaction went through the Zodiac Roles Modifier, which checked it against the wallet's configured permissions and allowed it. He suggested the most likely cause was that a role member's key had been compromised - and that the wallet's custom policy permitted the transfer, probably unintentionally.

The incident is a reminder that wallet security isn't only about whether a contract has a bug. It's also about who can authorize transactions, what those authorizations allow, and whether the rules still match what the wallet is actually meant to do. This article looks at what's publicly known about the loss, how role-based permissions fit into the transaction path, what the available evidence supports (and doesn't), and how teams can review their own authorization controls without assuming one setup fits every wallet.

What happened in the Safe wallet incident

Blockchain security firm PeckShield flagged a suspected theft, citing onchain investigator Specter. Records reviewed by Cointelegraph said that on Monday, approximately 4.39 million USDC and 1.72 million syrupUSDC were transferred from a Safe wallet in a single transaction. The reported value of the assets was about $6.45 million.

The wallet's owner hasn't been publicly identified. The incident record doesn't show who controlled the wallet, how the suspected attacker got access to a key, or whether the funds can be recovered. The transfer itself is visible onchain - but the authorization circumstances behind it are not fully public.

Johannes Moormann, Zodiac's co-founder, said his team reviewed the transaction and confirmed it passed through the Zodiac Roles Modifier. The module evaluated the transfer under the wallet's configured policy and allowed it. Moormann said the transaction came from a key belonging to a role member with an established relationship, and that it was most likely the result of that key being compromised.

He also said Zodiac's contracts worked as intended and that the wallet's custom permissions policy allowed the transaction, almost certainly unintentionally. Moormann said the wallet operators controlled the keys and permissions, and that Zodiac didn't have access to them or help set up the wallet. Zodiac had not been in contact with the affected team at the time of the report.

That distinction matters. Saying "the contract followed its rules" isn't the same as saying "the setup was safe" or "the loss was harmless." Systems can enforce their rules exactly as written and still allow an outcome the operators never meant to permit. In incident reviews, "the contract behaved correctly" and "the wallet was configured securely" are separate questions.

Key takeaways

  • The reported loss came from a single transfer of USDC and syrupUSDC from a Safe wallet.
  • Zodiac's co-founder pointed to a likely compromised role-member key and a custom policy that allowed the transfer.
  • The report's public material doesn't establish who owned the wallet or how the key was compromised.
  • Similar fund flows elsewhere don't prove a shared attacker on their own.

How a permission module can allow a harmful transaction

Safe wallets use smart contracts to manage assets and authorization. The Zodiac Roles Modifier adds another step in that flow: wallet operators define roles and specify which transactions those roles can perform. The module then checks a transaction against the configured permissions.

Done well, this can make day-to-day operations easier. A role might be able to execute a limited set of routine actions without using the same approval path for every high-value treasury transfer. That can reduce friction for teams that do frequent transfers or protocol operations. But the real security benefit depends on how narrowly the role is scoped and how strongly the key is protected.

A permission module doesn't make a judgment call about whether a transaction makes sense in context. It just enforces the rules it was given. If those rules allow a role to transfer a specific asset - or call a wider function than the operators intended - the module may permit an action that's damaging. In other words: the control is only as restrictive as the policy behind it.

That leads to two common failure paths that can show up separately or together:

  • Credential compromise: An unauthorized party gets a key that the system accepts as belonging to an authorized role member.
  • Overbroad permissions: The role's permissions cover more actions, assets, amounts, or destinations than the operators planned to expose.

These can stack. A stolen key paired with tight permissions may limit how much damage is possible. But a role with broad permissions can turn one compromised credential into a much bigger incident. On the other hand, even a restrictive policy can't fully compensate for weak key custody - especially if the allowed actions include high-impact operations.

There's a trade-off, too. Narrow, transaction-specific controls can reduce the blast radius, but they may add approval steps and slow operations. Broader permissions are easier to work with, but they increase the cost of a compromised key or configuration mistake. The right balance depends on what the wallet is for, how often it transacts, what it holds, and whether the team can monitor and respond quickly.

So the operational question isn't just "Does this wallet use a permission module?" It's: If a role key is compromised, what can that role do - and how fast would the team notice? That tests the authorization design, not the mere presence of a security component.

What the evidence says about the suspected attack

The public record supports a limited set of conclusions. The reported transfer happened onchain, with the funds leaving the wallet in a single transaction. Zodiac's co-founder said that transaction passed the Roles Modifier's configured checks. His explanation points to a likely compromised key and a policy that allowed the transfer. What it doesn't do is identify the attacker or provide a confirmed account of how the key was obtained.

After the funds left the wallet, PeckShield said the attacker used about $813,000 to buy Monero-linked tokens on Hyperliquid. Separately, Bitquery traced funds through multiple addresses to Hyperliquid, where five accounts reportedly bought about $1 million in XMR1 tokens. These figures describe analysis of later flows and purchases; they shouldn't be treated as a complete accounting of the original wallet's loss or as proof that every dollar followed the same path.

Bitquery also noted similarities between these movements and fund flows following a separate Lighter incident, including a 160 ETH transfer and XMR1 purchases. But Bitquery said it found no transfers connecting the two wallet groups and couldn't determine whether the same actor was involved. Similar transaction patterns can be useful leads, but they don't automatically establish attribution - different actors can use the same exchanges, tokens, or laundering tactics.

For wallet operators, this is the practical lesson: separate what's observed, what's assessed, and what's unresolved.

  • Observed: the asset transfer and subsequent onchain movements analysts reported.
  • Assessment: the explanation that a role-member key was likely compromised and the policy allowed the transaction.
  • Unresolved: who controlled the wallet, how the key was obtained, whether a related incident involved the same actor, and whether any funds can be recovered.

That separation helps teams avoid building controls around a story that hasn't been proven. If the key was compromised through endpoint malware, the response looks different than it would for an exposed recovery phrase, a compromised signing device, or a permissions error. Until the entry point is verified, teams should investigate multiple plausible paths instead of treating the first explanation as settled.

How wallet teams can reduce similar risk

A useful review starts with the transaction path, not a checklist of security products. Map each step from the person or service initiating an action, to the key that signs it, the role that key belongs to, the permission check, and the final wallet execution. For each step, ask what could go wrong and what evidence would confirm it.

Review permissions against real operations

For each role, document what it can do: which assets and contracts it can touch, any amount or frequency limits, and whether destinations are restricted. Then compare the policy to the role's actual job. If a role is meant for routine operations but can move a material share of treasury assets to arbitrary destinations, the authorization is broader than the operational need.

Test the policy with specific scenarios. What's the maximum value a compromised role key could move? Can it approve or interact with a new contract? Could it act repeatedly before operators intervene? If a permission review can't answer these questions, it's not yet doing much for risk management.

Treat role keys like real production credentials

Role keys should have custody controls that match the authority they carry. That may mean hardware-backed signing, restricted access to the signing environment, separation between operational and administrative roles, and a process for quickly removing access when team members or devices change. The point is simple: match custody to impact. A low-friction key isn't appropriate just because the role is used for routine transactions.

Add limits and detection where the workflow allows

Where possible, add transaction limits, destination allowlists, time delays, or extra approval steps for actions that exceed normal operating patterns. These controls can slow legitimate work, so they shouldn't be applied blindly to every transaction - apply them where the risk justifies the friction.

Monitoring should be tied to the policy, not added as noise. Alert on unusual assets, destinations, values, role usage, and changes to permission settings. An alert only helps if someone owns it, can assess it quickly, and has a tested way to pause or revoke authority. A notification nobody checks isn't a control.

Rehearse the response before an incident

Write down who can disable roles, change permissions, move unaffected assets, contact counterparties, and coordinate the investigation. Then test those steps with the people who would actually do them. In an incident, unclear authority can cost time, and changing permissions without understanding dependencies can disrupt legitimate operations.

The goal isn't to eliminate risk or make every transaction painful. It's to bound the damage from a single compromised credential, make abnormal activity visible, and give the team a response path that works under pressure. Revisit the controls when wallet responsibilities, asset balances, team membership, or transaction patterns change - not only after a loss.

Conclusion: review the whole authorization chain

The reported Safe wallet loss shows why smart contract correctness is only one piece of wallet security. A valid permission check can still authorize a damaging transaction if a role key is compromised and the policy is broader than intended. For teams managing onchain assets, the practical move is to map authority end to end, limit what each role can do, protect those credentials, monitor meaningful changes, and rehearse how to respond. Security holds when these controls work together - not when any single component is treated as enough.

Frequently Asked Questions

Zodiac’s co-founder said the transaction passed the Roles Modifier’s configured checks and that he saw no vulnerability in the module. The reported explanation was a likely compromised role-member key combined with a permissive wallet policy. The public information does not independently establish every detail of the incident.
The reported transfer included approximately 4.39 million USDC and 1.72 million syrupUSDC, valued at about $6.45 million. The transfer was reported as a single onchain transaction. The wallet owner has not been publicly identified in the available account.
The module lets Safe users assign permissions to roles for specific transactions. It checks an action against the policy configured by the wallet operators. Its protection therefore depends on the scope and accuracy of those permissions, as well as the security of the keys authorized to use them.
Bitquery identified similarities in some fund movements, including transfers involving ETH and purchases of XMR1. However, it reported finding no transfers connecting the wallet groups and could not determine whether the same actor was involved. Similar patterns alone do not establish a link.

Related articles