Google DeepMind AI Finds Critical Weaknesses in Legacy RSA-1024 Encryption

By Hector Herrera | November 7, 2026 | Security

Google DeepMind researchers working alongside human cryptanalysis experts have discovered two significant weaknesses in legacy RSA-1024 cryptosystems — the encryption standard that still secures a substantial portion of older internet infrastructure, industrial control systems, and financial networks. The findings don't break RSA-1024 entirely, but they dramatically lower the cost and effort required to do so, adding urgent pressure on organizations that have delayed migration to stronger encryption.

RSA (Rivest–Shamir–Adleman) is a public-key cryptosystem that secures data by relying on the computational difficulty of factoring very large numbers. The larger the key size, the harder the factoring problem. RSA-1024 uses 1,024-bit keys, which have been considered deprecated by security authorities since 2013 but remain in widespread use across legacy systems.

What Was Found

According to reporting from DEV Community's AI Daily Digest, the DeepMind team identified two distinct weaknesses:

1. Signature forgery efficiency improved roughly 1,600-fold. The AI-assisted analysis found structural properties in RSA-1024 key generation that make it possible to forge cryptographic signatures far more efficiently than previously known methods allowed. Signature forgery is the basis for impersonating trusted servers, signing malicious software as legitimate, or forging digital documents.

2. Batch factoring cost fell approximately 30-fold. When an attacker has access to a large collection of RSA-1024 keys — as occurs in large-scale network scanning — the marginal cost of factoring each additional key dropped dramatically. This is relevant to nation-state actors and sophisticated threat groups that can amass key sets at scale.

Neither finding means that RSA-1024 can be broken in real time with consumer hardware. But both findings mean that adversaries with significant computational resources — large server farms, cloud compute budgets, or access to specialized hardware — now face substantially lower barriers to attacking RSA-1024 infrastructure than was understood before this research.

Who Is Still Running RSA-1024

Despite years of deprecation warnings from the U.S. National Institute of Standards and Technology (NIST), the European Union Agency for Cybersecurity (ENISA), and industry standards bodies, RSA-1024 remains embedded in:

• Older industrial control systems (manufacturing, utilities, water treatment) where firmware upgrades are operationally complex
• Legacy financial infrastructure in emerging markets where system replacement cycles are long
• Embedded devices — including some medical equipment, point-of-sale terminals, and building management systems — that cannot be easily patched
• Internal enterprise systems that were not included in public-facing TLS upgrade programs

The persistence of RSA-1024 is not negligence in every case. Many of these systems are physically difficult to update, run on hardware that cannot support newer cryptographic libraries, or are part of regulatory-approved configurations that require expensive re-certification after any change.

Why AI Changed the Analysis

What makes this research notable is the method. Human cryptanalysts have studied RSA weaknesses for decades, and the mathematical properties of RSA-1024 were believed to be well-understood. What the DeepMind team demonstrated is that AI-assisted analysis — specifically, the ability to rapidly search large mathematical solution spaces and identify non-obvious structural patterns — can surface weaknesses that decades of human cryptanalysis missed.

This is a consistent theme in AI-assisted security research: not that AI "hacks" systems in the action-movie sense, but that AI enables researchers (and attackers) to find the gaps between what we thought we understood and what is actually there.

What Organizations Should Do

The practical guidance from security authorities has not changed, but the urgency has:

• Migrate away from RSA-1024 immediately if you have not already. The target is RSA-3072 or higher for symmetric-equivalent security, with post-quantum cryptography (PQC) algorithms — standardized by NIST in 2024 — as the longer-term destination.
• Audit legacy systems for RSA-1024 key usage, particularly in environments that are rarely included in standard vulnerability scans (OT/ICS networks, embedded systems, internal certificate authorities).
• Do not wait for active exploitation to trigger migration. The 1,600x improvement in signature forgery efficiency in particular suggests that well-resourced attackers may already be ahead of public disclosure.

What to Watch

Watch for CISA and NIST to issue updated emergency guidance citing the DeepMind findings within the next 30 days. The research also strengthens the case for mandatory PQC migration timelines in regulated industries — healthcare, finance, and critical infrastructure — where regulatory agencies have been reluctant to impose hard deadlines. This finding may change that calculus.

Source: DEV Community AI Daily Digest, November 7, 2026