As federal agencies begin transitioning their most sensitive systems to new forms of encryption following a White House directive in June, the recent release of advanced theoretical mathematics results by OpenAI has raised concerns about the security of even proposed post-quantum encryption methods against increasingly powerful AI models.
Encryption is essential for maintaining online data security; it involves applying complex mathematical algorithms to obscure data from unauthorized access. However, emerging mathematical techniques could potentially allow attackers to decode encrypted information without a key—akin to a safecracker deducing a combination by listening to the clicks within a lock. Should current encryption methods be compromised, vast portions of the internet would become vulnerable to exploitation.
Consequently, post-quantum encryption strategies may require more robust, multi-layered protections—and accelerated timelines beyond what the White House originally anticipated—to safeguard data that has yet to be collected.
Presently, encryption remains effective because there are standardized methods for authorized parties to apply mathematical operations to decrypt messages, while unauthorized decryption through brute-force attacks remains computationally infeasible. However, current encryption algorithms lack absolute theoretical proof of invulnerability; they are trusted primarily due to the absence of successful breaches thus far. Any advancement in mathematical techniques that simplifies the decryption process could jeopardize the foundational security infrastructure of the internet.
AI systems are generating mathematical breakthroughs at an unprecedented pace. In May, an OpenAI model disproved the Erdős unit distance conjecture, a renowned problem in discrete geometry. Later in September, the company revealed its AI had developed a solution to the Navier-Stokes equations—one of the seven Millennium Prize Problems. These complex mathematical challenges rank among the most significant unresolved issues in mathematics, often representing career-defining achievements for researchers.
“It just solved 400 problems, any of which, if a human had solved it, that could have been the highlight of their career, that could have been their tenure case,” said Scott Aaronson, author of Quantum Computing Since Democritus and Chair of Computer Science at UT Austin, in an interview.
Omer Reingold, a computer scientist and co-winner of the Gödel Prize in 2009, expressed concern prior to the latest release: “The world of mathematics, including theoretical computer science, is in turmoil. Even the Millennium Prize Problems…are no longer beyond the reach of LLMs.”
Aaronson observed that the OpenAI papers contained results across nearly every mathematical field—except cryptography. “And that can’t possibly be an accident,” he noted.
Justin Drake, a security researcher at the Ethereum Foundation, suspects deliberate censorship of encryption-related findings. “I’ve witnessed first-hand the U.S. government censoring academic quantum cryptanalysis results. Backroom interventionism is my base case,” he stated.
Regardless of whether OpenAI intentionally withheld encryption-related discoveries, the wave of new findings has intensified fears that AI systems could uncover vulnerabilities in existing encryption protocols. The cryptocurrency sector, which depends heavily on encryption to protect its digital assets, has been especially alarmed by these developments.
Cryptocurrencies operate in part by exposing certain account information during transactions. On X (formerly Twitter), Drake issued a warning: “Today I call upon the blockchain industry to calmly begin planning for ‘bunker mode’. My personal recommendation is to initiate a controlled mass migration of assets to fresh addresses.”
A “fresh address” refers to one that has never previously been used to send funds, thereby minimizing potential attack vectors for future threats.
While some cybersecurity experts remain skeptical about AI’s immediate impact on cryptographic security, others acknowledge broader concerns. Bruce Schneier, renowned security expert and author of the bestselling book Applied Cryptography, emphasized the cautious nature of cryptographers: “They build enormous buffers into secure systems. The real challenge lies in implementation flaws and system vulnerabilities—not necessarily breakthrough-level mathematics.”
“Computer security today is fundamentally flawed. Vulnerabilities in code, system intrusions, and implementation errors pose the greatest risks,” Schneier added. “Math remains our strongest defense mechanism in the realm of cybersecurity.”
Emily Wenger, Assistant Professor of Electrical and Computer Engineering at Duke University, who has employed machine learning techniques to evaluate post-quantum encryption resilience, stated: “I don’t believe large language models will break all encryption.” She highlighted two distinct pathways for compromising cryptographic systems: proving mathematical weaknesses or exploiting flawed implementations. “We’ve seen historically that even theoretically sound algorithms can be undermined through poor execution,” she cautioned.
Aaronson warned of the potentially catastrophic implications of undermining widely adopted encryption schemes like RSA. “If such a foundational algorithm were broken, it would create systemic failure across internet infrastructure—the way we’ve built digital trust globally would collapse,” he explained.
Dan Boneh, a cryptography expert and professor at Stanford University, reinforced this sentiment in a recent video discussion: “Loss of current encryption standards would impact the entire internet and global digital ecosystem. Critical sectors—including banking, healthcare, and government communications—would face severe disruption.”
Nevertheless, some experts view these challenges as opportunities for strengthening cryptographic frameworks. Boneh noted: “Historically, deploying new cryptographic systems required lengthy peer review cycles—often spanning decades. Today’s AI-driven acceleration allows rapid testing and validation. By simulating countless attack attempts within short periods, we gain unprecedented confidence in our security assumptions.”
Schneier agreed: “This represents an evolving arms race—but perhaps uniquely, AI might tip the balance toward defenders. For instance, Microsoft recently leveraged AI to identify and patch hundreds of software vulnerabilities during a major update cycle. Such proactive defenses counteract threats at machine-speed scales.”
Despite potential benefits, transitioning encryption infrastructure carries substantial costs. Organizations are pursuing “crypto agility”—the capability to rapidly adapt or replace encryption mechanisms—as a strategic response to uncertainty. While small businesses may struggle with the financial burden of overhauling systems, defense contractors face even greater challenges, as adversaries have likely already harvested and stored sensitive data encrypted under outdated protocols.
Governments worldwide have long practiced “harvest now, decrypt later” intelligence gathering—collecting and retaining encrypted communications anticipating future decryption capabilities. With AI advancing faster than expected, this threat model becomes increasingly urgent.
“Assume foreign and domestic intelligence agencies are actively collecting your data today—that assumption grows more valid if you believe AI breakthroughs could expose encrypted transmissions soon,” Aaronson advised.
For institutions managing classified or restricted information, the convergence of AI advancement and cryptographic evolution introduces another pressing timeline in the transition toward post-quantum readiness—an effort still in early stages for many organizations.
Federal mandates outline a phased migration to post-quantum cryptography between 2030 and 2035 under Executive Order 14412 and OMB Memo M-26-15. National Security Agency guidelines establish parallel timelines for defense-related systems. Meanwhile, contractors must comply with updated NIST cryptographic standards by 2030.
Yet emerging mathematical discoveries tied to AI could necessitate revised strategies altogether. As Aaronson observed: “Classical versus post-quantum encryption failures aren’t mutually exclusive—they operate independently. Either scenario demands preparedness.”
Experts widely advocate adopting flexible architectures capable of adapting to evolving threats. Crypto agility—the ability to seamlessly integrate or swap encryption methodologies—emerges as a recommended best practice regardless of specific breakthroughs.
“Crypto agility should begin now—not deferred to future planning cycles,” Schneier urged. “Organizations lagging behind should prioritize establishing modular, scalable encryption ecosystems immediately.”
Matthew Green, Associate Professor at Johns Hopkins’ Information Security Institute and author of the blog A Few Thoughts on Cryptographic Engineering, succinctly captured the dilemma facing technologists: “Predicting the trajectory of AI-powered cryptanalysis defies conventional forecasting. Yet history shows betting against computational models rarely yields favorable outcomes.”