Get your free personalized podcast brief

We scan new podcasts and send you the top 5 insights daily.

Cryptographers traditionally waited a decade for human researchers to study and build confidence in hardness assumptions. If frontier AI labs spend millions of dollars in compute trying to break these schemes and fail, it provides far greater conviction much faster than traditional academic scrutiny ever could.

Related Insights

Friedberg observes that OpenAI released hundreds of advanced mathematical solutions while conspicuously omitting proofs concerning cryptography. Because mathematical advances occurred across multiple complex domains, the absence of public key cryptography research suggests breakthrough findings were intentionally withheld for national security.

The security of software is increasingly determined by how much AI compute was spent trying to break it. Companies use diverse AI agents to autonomously attack their own code. The amount of money spent on APIs from labs like Anthropic to find vulnerabilities is becoming the best indicator of a system's robustness.

Quantum computing threats were mathematically formalized decades in advance through known algorithms like Shor's algorithm. In contrast, AI-accelerated mathematical advances could uncover novel algorithms on commodity hardware without giving cryptographers advance warning of the threat's shape.

While crypto firms seek access to next-gen AI for security testing, the real insight is that current-generation models are already proving superior to human auditors. For example, crypto custodian Fireblocks found that an existing Anthropic model detected critical vulnerabilities that multiple professional security audit firms had missed.

Cryptographers cannot mathematically prove that their underlying problems are intractable because doing so requires resolving the long-standing P versus NP problem. Consequently, cryptography relies on assumptions that remain trustworthy only because decades of attempts by human researchers have failed to find efficient attack algorithms.

AI models succeed at disproving conjectures because testing counterexamples can be partitioned across thousands of parallel agents. While human researchers are constrained by expert consensus and get discouraged by exotic, low-probability constructions, AI exhaustively checks unconventional paths without cognitive bias.

Human-led jailbreaking efforts will become obsolete. The next frontier involves one AI using massive compute (hundreds of thousands of GPU hours) to systematically test every conceivable input sequence against another AI. This industrializes the search for vulnerabilities, turning security into a computational arms race where sufficient compute guarantees a breach.

The same AI power that solves complex math problems can be applied to formally verify software, a long-sought-after goal in computer science that was previously intractable for humans. This could create a new paradigm for building secure and bug-free systems.

Quantum algorithms specifically threaten systems relying on discrete logarithms and RSA factoring. In contrast, advanced AI models apply general mathematical reasoning that could theoretically discover weaknesses across classical systems and modern post-quantum candidate schemes like lattices and hash-based architectures alike.

AI now generates complex scientific derivations faster than humans can validate them. For a recent quantum gravity paper, the AI produced the core results in days, but human collaborators spent three weeks just checking the work, shifting the research bottleneck from discovery to verification.