DeepSafe

DeepSafe is excited to announce our $3m Seed Round with @AntalphaGlobal @ViabtcCapital @capital_spark @CogitentV @ShardingCapital @Mason_eagle @Gate @SatoshiLab_HK @CKBEcoFund

The core decentralized validation technology of DeepSafe has been accepted by the top international cryptographic academic journal IEEE TIFS, with the document ID being 9903072.

Currently, the DeepSafe network has processed nearly 120 million transaction validations, and the number of active accounts on the network exceeds 2.65 million.

Time to make a better Trust Layer now.

Running key management inside SGX keeps a device's data confidential and lets it detect tampering. It doesn't tell the device which version of that data is current. A Trail of Bits post on threshold signing in TEEs puts the gap in one line: "Authenticated encryption provides confidentiality and integrity but not freshness."

CRVA runs MPC inside SGX, and its docs name a version of this problem for its own devices: the Database Backup Attack.

Devices act on chain events and chain state, which they read through a node their owner configures. The docs walk through how an owner with several devices could use that: back up the devices' databases every epoch; wait until identities are revealed after the next handover; and, if those devices turn out to hold enough shares to rebuild a committee's key, "send fake events and fake storage" to the backups so the devices hand their shares to devices the owner created.

Device data is encrypted by SGX, and nothing in those steps requires breaking that. As we read it, what the attack replaces is the device's view of which chain state is current.

The defense in the docs is a vote. Each epoch, the snapshot of working devices goes to the previous epoch's CourtMembers. With more than half in favour it is confirmed, and its devices become the current CourtMembers; without that, the previous members stay on. The Ring-VRF candidates then go to the current CourtMembers. If the candidates do not get more than half of the current CourtMembers, the candidates are cleared and no key generation or handover runs. To attack this way, the docs say, an attacker would need more than half of that epoch's CourtMembers.

An enclave can tell whether its data was altered. For the state that decides who receives key shares, CRVA doesn't leave the question of what is current to any single device.

blog.trailofbits.com/2026/09/25/don…
docs.deepsafe.network/introduction/c…

#DeepSafe #CRVA #Web3Security

A verification layer is only as useful as the precision of what it claims to check. For one of its components, DeepSafe spells that out.

The SPV service runs blockchain light clients inside SGX and provides transaction verification to the monitor. Its stated purpose, in the docs: "preventing malicious signatures caused by fraudulent information sources."

On the Bitcoin side the docs describe the familiar construction: block headers, Merkle proofs, confirming "a transaction's inclusion in a specific block."

There is a second question, about the checker rather than the chain: how do you know which program produced the answer? The service registers through remote attestation and its enclave measurement is validated on chain, so that "only programs with the specific, pre-approved code can pass verification." In local mode each monitor is paired with its own SPV, to "ensure decentralization of verification sources."

That is a narrow question, answered carefully.

What it does not answer is what the transaction means. An inclusion proof establishes that a transaction is included in a particular block. It establishes nothing about which account an application should credit, or what an instruction attached to it authorizes. Those follow from the application's own rules, and the rules differ from one bridge to the next.

The two are not the same kind of question. Whether a transaction is included in a given block is a question about the chain, and it is worth solving once, at the layer. What that transaction means is a question about one system's rules, and the answer has to come from those rules.

A check is more useful when you can say where it stops.

Sources:

docs.deepsafe.network/introduction/s…

#DeepSafe #CRVA #Web3Security

Weekend puzzle.

A key is split into 5 shares. Any 3 of them can sign.

One share leaks in January. Another in June. A third in November. Someone ends up holding all three.

In November, can they sign?

A. No. The leaks were months apart.
B. Yes, unless the shares were refreshed in between.
C. No. Each leak was patched when it was found.

On 5 October, the Ethereum Foundation's Trillion Dollar Security initiative published a post on native transaction assertions, with EIP-7906 as one possible design. For the Hegotá upgrade, EIP-7906 has advanced to "Considered for Inclusion" but is not yet confirmed.

The idea: a transaction carries a rule about its own outcome, signed along with its actions. After the actions run, read-only code compares starting and final values against that rule: a minimum amount received, no new approvals, an account's control logic left unchanged. If the rule fails, the actions revert.

That puts the check after the actions have run, where the full set of net state changes can be inspected. The post is direct about two conditions it still depends on.

The first is where the rule comes from. "A compromised frontend can write an assertion that permits the attack." A useful rule, in the post's words, "must come from independently approved user intent, a standing account policy or protocol logic that the compromised component cannot change."

The second is whether the rule is there at all. "EIP-7906 does not require transactions to include an assertion, so whoever builds the transaction can leave it out." A wallet has to include it in every frame transaction it builds (the transaction type EIP-7906 builds on, defined in EIP-8141); a protected function has to reject ordinary transactions and frame transactions that arrive without the exact assertion it expects.

What the check protects depends on two things outside the check itself: who wrote the rule, and whether anyone can leave it out.

The same two questions apply to any verification layer, including ours: whose rules it checks against, and whether it can be left out.

blog.ethereum.org/en/2026/10/05/…

#DeepSafe #Web3Security