Home · Learn · Announcement, November 2, 2026

A large exchange theft was a signing-path failure. Here is how that path looks on JIL.

On October 29, 2026, about $6.3 billion left one exchange's hot wallets across roughly ten blockchains. The chains worked as designed. The signing path did not. The useful question for any settlement system is what stands between a compromised signer and a transfer, and which of those controls are running today.

What was reported

Public reports put the loss at about $6.3 billion, taken in roughly twenty to thirty minutes from hot wallets on about ten chains. The exchange named in those reports, Binance, traced the breach to a compromised engineer laptop. The reports describe the attack as a compromise of the exchange's internal signing infrastructure, not a flaw in any of the chains. Figures and details differ between outlets, and a full post-mortem has not been published. The sources are listed at the end of this page. This page names no attacker and makes no attribution.

The pattern matters more than the number. Once the attackers could make the signing path produce the signatures they wanted, each chain executed them as valid transfers. A valid signature was the only thing the chains asked for.

The design question

If an attacker controls a signer, what else has to say yes before value moves? On the JIL Layer-1 the answer is a second decision made outside the signer. A signature is necessary and is not sufficient. ATCE, the Adaptive Trust, Compliance and Execution Engine, rules on each transfer inside consensus, and a review step can hold an act that the rules would have allowed.

The dated status of each ATCE capability is kept on the ATCE status ledger and the site-wide status ledger.

Each control against this attack path

ControlWhat it does against a compromised signerStatus
Deterministic rules in consensus (ATCE)The chain rules on the sender and the recipient and denies a transfer that policy forbids, even when the transaction carries a valid signature. The ruling has two outcomes, allow or deny, with a reason code.Built. Measured 2026-11-12, the every-transfer rule is not in the validator image created 2026-08-21. That image can deny a crossing.
Hold-only reviewA review of the specific act can deny it even when every rule passed. A held act is a deny, and a person decides what happens next outside the review. The review cannot release an act the rules denied, and a hostile input cannot override the rules.Built. The hold-only review was not confirmed on the validator image created 2026-08-21.
Hybrid attestation seals (Ed25519 and ML-DSA-65)Protects the evidence of what was attested. It does not protect a signing host from compromise.Live. On chain, the Ed25519 half is verified today.
Post-quantum validator signingNot relevant to this theft, which used ordinary valid signatures, not a quantum attack. Post-quantum signing protects against a different future threat.Designed. Validator signing keys are Ed25519 software keys today.
Default service keyA 2-of-3 threshold. All three shares are held server-side by JIL, and signing reconstructs the key in memory. There is no user-held shard. A compromised signing host is still a target.Live. Threshold signing that never reconstructs the key is Designed.
Write-once record of each rulingPreserves the full record under the ruling's SHA-256 commitment so a transfer can be re-read afterward. It records what happened and does not prevent a transfer.Built. The record writer was not confirmed on the validator image created 2026-08-21. The production store is Designed. Retention is 10 years from creation.

What this means for the signing path

JIL places a rule engine and a hold-only review between a signature and a settled transfer. A stolen signature alone is not enough to move value that policy forbids. That is the part of the system aimed squarely at this kind of incident.

The design does not remove the signing host as a target. A compromised signing service is still a target. We would rather say that plainly than claim the attack path is closed.

What we do not claim

  • That this kind of theft cannot happen on a JIL chain.
  • That post-quantum signatures would have prevented it. They would not have.
  • That the company cannot sign on the default key.
  • That the details of the incident are final. They come from public reports and may change.

Sources

This page is informational. It is not security, legal, or investment advice.

Keep reading

The status ledger dates every ATCE capability. The post-quantum page states what the seals cover.