Google DeepMind says supervised AI agents found two weaknesses in legacy RSA-1024 encryption
The research suggests older 1024-bit RSA systems can be broken with much less computing power than previously assumed.
Google DeepMind has published research showing that AI agents, working under human supervision, can greatly speed up cryptanalysis, the work of stress-testing encryption systems to find where they fail, OfficeChai reports. According to the team, its agents found two significant weaknesses in legacy 1024-bit RSA cryptosystems. The findings suggest these older systems can be compromised with far less computing power than previously assumed. Dj Dvijotham and Abhradeep Thakurta led the research, working with Nadia Heninger of UC San Diego and Inria and Emmanuel Thomé of Inria. RSA's security depends on an asymmetry: multiplying two large prime numbers is easy, but factoring the result back into those primes is extremely hard. That difficulty underpins encrypted communications, digital signatures and much of the internet's security infrastructure. OfficeChai reports that DeepMind was careful to say the result is not a reason to panic, because the work concerns the legacy 1024-bit configuration. The study is part of a broader move by AI labs to apply agent systems to open research problems rather than narrow benchmark tasks. The findings matter most to organizations that still run older RSA-1024 deployments, which now have more reason to move to stronger keys.