A team of researchers at UC San Diego and France’s INRIA has pulled off something that sounds almost like a magic trick: forging valid RSA signatures on a 1,024-bit key locked inside a hardware security module, without ever removing that key from the device. The finding, detailed in a paper submitted to the IACR Cryptology ePrint Archive on September 20, has drawn attention across the cybersecurity world not because it breaks the internet overnight, but because it shows a working RSA signature attack against a device designed specifically to prevent exactly this kind of forgery.
Summary
Key takeaways
- Researchers at UC San Diego and INRIA forged RSA signatures on a 1,024-bit key inside a hardware security module without extracting the key.
- The attack needed roughly 4 billion signing requests and about 1,380 CPU core-years of computation.
- It exploits an unpadded RSA signature oracle that most modern, properly configured RSA systems do not expose.
- Bitcoin and Ethereum rely on elliptic-curve signatures (ECDSA), so they are not directly affected.
- Google has set 2029 as its internal deadline to shift toward post-quantum cryptography.
Details of the RSA Signature Forgery Attack
The short answer to how this happened: the researchers didn’t crack RSA’s underlying math, they tricked the machine into helping them fake a signature. They treated the hardware security module, or HSM, like a vault that refuses to open but will stamp any blank slip of paper slid under the door. Ask it to stamp enough slips, and eventually you can learn to reproduce that stamp yourself, without ever touching what’s inside the vault.
Attack Methodology and Conditions
To make that trick work, the team had to switch off the HSM’s FIPS mode, a certified security setting meant to enforce proper formatting on everything the device signs. With FIPS mode disabled, the module agreed to sign raw, unformatted numbers instead of properly padded data. That gap is what created what cryptographers call a signature oracle: a system that, if asked the right way over and over, leaks enough information to reconstruct a forgery. The researchers used a test key they controlled for the demonstration, rather than exploiting a live, production key.
Computational Efforts Required
This wasn’t a quick job. The attack demanded roughly 4 billion signing requests, on the order of 2^32, sent to the module, followed by about 1,380 CPU core-years of number crunching to turn those responses into a working forged signature. Even by cryptographic-research standards, that’s a heavy computational lift, and it’s a big part of why the paper frames this as a proof of concept rather than an active real-world threat.
Scope and Limitations of the RSA Signature Attack
This RSA signature attack does not open a door into properly configured modern systems. Standard RSA signing applies padding, a scrambling and formatting step using schemes like PKCS#1 v1.5 or PSS, before any signing math happens. That padding step is precisely what closes off the exploitable oracle the researchers relied on. In other words, most RSA deployments running today, the ones that follow standard padding practices, simply don’t hand attackers the raw material this attack needs.
Impact on Modern Cryptographic Systems
Because the attack depends on disabling a certified security mode and working with unpadded signatures, it targets a narrow and largely avoidable configuration rather than RSA itself as a whole. That distinction matters for anyone trying to gauge real-world risk: the vulnerability lives in a specific setup choice, not in the core RSA algorithm that underpins countless certificates, software signatures, and authentication systems worldwide.
Comparison with Bitcoin and Ethereum Signature Schemes
For crypto users specifically, there’s a clean answer here: no, this doesn’t touch Bitcoin or Ethereum. Both networks sign transactions using elliptic-curve signatures, specifically ECDSA, an entirely different cryptographic approach from RSA. The paper’s claims are scoped to RSA and its padding behavior, so wallets and blockchains built on elliptic-curve schemes sit outside this particular attack’s reach.
Use of Blind Signatures in Privacy Technologies
Where things get more interesting is with systems that deliberately expose an RSA signing oracle on purpose. RSA-based blind signatures let a server sign a piece of data without ever seeing its contents, a technique put to use in privacy tools such as Privacy Pass. Cloudflare has noted that Apple uses a version of Privacy Pass so users can prove they’ve passed a check, like a CAPTCHA, without giving up their identity. Blind signatures aren’t new; cryptographer David Chaum built the technique into DigiCash back in 1989. But their design, by nature, involves the kind of oracle behavior this new research probes, which is why the paper’s authors flag privacy-preserving RSA applications as worth a closer look.
Broader Implications for Cryptographic Security
The real significance of this research isn’t a live threat to your wallet or your browser today. It’s what the researchers themselves call classical evidence for moving away from RSA as part of the broader shift toward post-quantum cryptography. Every demonstration that chips away at confidence in a decades-old standard, even one bounded by heavy computational costs and narrow configuration requirements, adds weight to the argument that the industry needs to keep migrating toward algorithms built to survive future threats.
Significance for Post-Quantum Cryptography Transition
RSA itself, created in 1977 by Ron Rivest, Adi Shamir, and Leonard Adelman, has held up for nearly five decades largely because factoring large numbers remains computationally brutal for classical computers. Notably, this new attack doesn’t factor anything at all. It sidesteps that hard problem entirely and instead exploits an operational weakness in how signatures were formatted. That’s a different kind of warning sign, one that speaks less to RSA’s math and more to how fragile real-world implementations can be under the wrong conditions.
Long-Term Threats from Quantum Computing and Google’s 2029 Plan
The bigger long-term concern remains quantum computing, not classical attacks like this one. Estimates vary widely on how many qubits it would take to run Shor’s algorithm against elliptic-curve schemes like ECDSA, but the underlying concern is treated as genuine for long-term security planning. Google has already set 2029 as its internal deadline to migrate its own systems toward post-quantum cryptography, a concrete signal of how seriously major infrastructure providers are taking the eventual quantum threat to both RSA and elliptic-curve cryptography.
Put together, this research doesn’t sound an alarm for today’s internet or today’s blockchains. It does something quieter but arguably more useful: it adds a documented, working example to the case for retiring RSA on a realistic timeline, well before quantum computers make the conversation urgent rather than precautionary.
FAQ
What is the nature of the RSA attack demonstrated by UC San Diego and INRIA researchers?
They forged RSA signatures on a 1,024-bit key inside a hardware security module without extracting the private key, by exploiting an unpadded RSA signature oracle.
Does this RSA attack affect Bitcoin and Ethereum security?
No, Bitcoin and Ethereum use elliptic-curve signatures (ECDSA), so this RSA-specific attack does not threaten them directly.
Why does the attack require disabling the HSM’s FIPS mode?
Because disabling FIPS mode allows the HSM to sign unformatted, un-padded numbers, which is necessary for the attack to exploit the RSA signature oracle.
What implications does this attack have for the future of cryptography?
It provides classical evidence supporting a transition away from RSA toward post-quantum cryptography, helping prepare cryptographic systems for evolving security threats, including those posed eventually by quantum computing.
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.




