Vitalik Explains the Cryptography Holy Grail: Diamond iO, a Planet-Scale Protocol That Makes Smart Contracts “Completely Invisible”

The “holy grail” technology in the cryptography field has brought major theoretical progress! In a new long-form article published today (28) on Vitalik Buterin’s official blog, Ethereum co-founder Vitalik delves deep into a brand-new cryptographic obfuscation protocol, “Diamond iO.” Through bold cryptographic assumptions and architectural innovation, the protocol successfully reduces the theoretical computation time for program obfuscation from the seemingly unreachable “galactic” scale down to “planetary” scale, bringing the ultimate technical gap for the internal logic of hidden code closer to real-world application.
(Background recap: Vitalik praises the Qwen model’s ability to recognize anonymity: the AI sees through his mathematical thought habits, and his prose camouflage completely fails)
(Additional context: Ethereum’s third major upgrade》Vitalik explains the Lean Ethereum roadmap: almost all key components will be replaced)

Table of contents

Toggle

• Breaking down layered stacking, slimming Diamond iO’s architecture
• Introducing new assumptions to block zeroizing attacks
• Future outlook: finding more efficient replacement solutions

In the world of blockchain and cryptography, how to execute a piece of code in a public environment while perfectly hiding its internal private keys and logic has long been regarded as the “holy grail” of technical innovation. On July 28, Taipei time, Ethereum co-founder Vitalik Buterin published the second article in his iO (Indistinguishability Obfuscation) series, unveiling the mysterious “Diamond iO” protocol to the public.

Compared with mainstream iO protocols that require “galactic” computation time—longer than the lifespan of the universe—Diamond iO, while still not feasible under today’s hardware conditions, has successfully reduced the complexity to “planetary” scale. Vitalik believes that with just a few rounds of key optimizations, this technology is highly likely to achieve genuine commercial deployment in certain specific use cases.

Breaking down layered stacking, slimming Diamond iO’s architecture

In the previous article, Vitalik introduced a more conservative branch of iO technology. Its operating principle is typically to perform fully homomorphic encryption (FHE) computations within attribute-based encryption (ABE), and then rely on a stack of layered protocol towers to ensure that the evaluator can only decrypt after correctly executing the evaluation. However, this architecture brings a devastating computational burden.

Diamond iO’s core breakthrough lies in two major modifications to the BGG+14 ABE scheme: introducing a completely different conditional FHE decryption mechanism, and a brand-new input encoding generation mechanism. This significantly reduces Diamond iO’s overall computational strength—roughly equivalent to Functional Encryption (FE)—rather than the endlessly complex stacking of the past.

In the workflow, the evaluator first obtains auxiliary items, the FHE encryption of the hidden input $z$, the public input bits $x$, and a low-norm FHE decryption key $s$. Next, the evaluator performs an FHE evaluation on the encoding to obtain the ciphertext’s BGG+ encoding, and then uses the property of “known-plaintext is multiplicatively and additively homomorphic” to carry out decryption computation. The protocol cleverly uses a “bucket-wise left shift” operator to scale the weights of different bits to avoid error blow-up, and finally extracts the result through a “trapdoor”—this trapdoor only allows computations that follow the correct circuit shape, ensuring the mechanism’s security.

Introducing new assumptions to block zeroizing attacks

To ensure rigor in the security proof, Diamond iO also makes adjustments on the function $f(x,z)$ side. To ensure the output is “jointly pseudorandom” for all inputs, the protocol employs a technique called “PROM bootstrap”: it makes the hidden input $z$ contain the true secret and the random PRF key, then computes a new function with an offset, $f'(x,z) = f(x,z_{\text{secret｝｝) \cdot q/2 + \text{PRF}(k,x)$). In addition, the protocol uses XOR hashing on both the output side and input side, successfully transforming “obfuscation only for pseudorandom functionality” into a general iO protocol.

In terms of security, the most controversial yet most innovative aspect of Diamond iO is its input encoding generation mechanism. To prevent the evaluator from leaking the secret via subtraction and similar methods, the protocol relies on two newer cryptographic assumptions: Evasive LWE proposed by Wee22, and All-product LWE unique to Diamond iO. These two assumptions effectively avoid the “zeroizing attacks” that have repeatedly defeated early obfuscation protocols, but Vitalik also admits that these assumptions are relatively new and carry some risk—future work will still require the cryptography community to invest more effort in security analysis and verification.

Future outlook: finding more efficient replacement solutions

Although the architecture has been greatly simplified, Diamond iO still faces many concrete performance bottlenecks. Vitalik points out that the current bottlenecks mainly come from the BGG+ evaluation overhead (the product of ABE and FHE), as well as the internal PRF (currently using a Goldreich PRG tree that deeply depends on the input length). At the same time, the protocol can currently only tolerate low-depth circuits to prevent excessive error accumulation, and it requires using a relatively large sub-exponential security parameter.

In response, Vitalik proposes multiple possible optimization directions for the future, including: using a more efficient PRF to replace the Goldreich tree, switching to more advanced FHE schemes (such as BGV/CKKS or packed GSW), further merging the ABE and FHE layers, and performing targeted optimizations for “iO full programs” specifically combined with STARK proofs.

At the end of the article, Vitalik summarizes that Diamond iO’s greatest value lies in its relatively simple structure and ease of analysis—developers don’t need to master all the complex constructions from the past 20 years to understand its core. He hopes this article will attract more researchers to invest in security analysis and performance optimization, ultimately enabling this deeply disruptive technology to truly enter our digital lives.