Listen to this Post
Introduction: A Historic Turning Point for Artificial Intelligence and Cryptography
Artificial intelligence has already transformed software development, cybersecurity, and scientific research, but its newest achievement may represent one of the biggest milestones yet. Anthropic has revealed that its advanced AI model, Claude Mythos Preview, has successfully conducted original cryptographic research with minimal human intervention, uncovering mathematical weaknesses that remained undiscovered after years of examination by some of the world’s leading cryptographers.
Unlike previous AI breakthroughs that focused on finding implementation bugs in software, this achievement goes much deeper. Claude Mythos analyzed the mathematics behind encryption algorithms themselves. That distinction is critical because implementation bugs can be fixed with software updates, while mathematical weaknesses can challenge the very foundations of cryptographic security.
Although none of the discoveries threaten
Anthropic Announces a Major Cryptographic Breakthrough
AI Moves Beyond Traditional Vulnerability Hunting
Anthropic announced that Claude Mythos Preview achieved two major research milestones almost entirely on its own.
The first involved discovering a significantly improved attack against HAWK, a post-quantum digital signature algorithm currently under review in the National Institute of Standards and Technology (NIST) standardization process.
The second achievement introduced a dramatically faster attack against a reduced-round version of the Advanced Encryption Standard (AES), improving previous research by approximately 200 to 800 times.
Importantly, neither discovery breaks any production encryption currently protecting users worldwide. Instead, both findings demonstrate the growing research capabilities of modern AI systems.
From Coding Errors to Mathematical Discovery
Claude Is No Longer Just Finding Software Bugs
When Claude Mythos was first introduced, it demonstrated impressive success by identifying vulnerabilities across numerous software projects, including several widely used cryptographic libraries.
Those vulnerabilities resulted from programming mistakes.
Developers had implemented otherwise secure algorithms incorrectly, creating opportunities for attackers.
This latest research represents something entirely different.
Instead of identifying coding mistakes, Claude examined the mathematical structures underlying cryptographic algorithms themselves and identified entirely new attack strategies.
That leap significantly raises the expectations for AI-assisted scientific research.
Understanding the HAWK Discovery
A Promising Post-Quantum Algorithm Faces New Questions
Among
HAWK remains one of the surviving candidates in NIST’s years-long effort to standardize cryptographic algorithms capable of resisting attacks from future quantum computers.
Claude identified an overlooked mathematical symmetry known as a nontrivial automorphism inside HAWK’s lattice structure.
That hidden symmetry enabled a substantially faster enumeration attack than researchers previously believed possible.
For the smallest HAWK-256 configuration, estimated attack complexity dropped dramatically.
Instead of requiring roughly:
2⁶⁴ operations
the new attack requires approximately:
2³⁸ operations
That reduction effectively cuts the intended security level in half.
Security Can Be Restored, But At a Cost
Bigger Keys Reduce Practical Advantages
Researchers noted that simply doubling
However, doing so introduces larger signatures and increased computational overhead.
One of
Increasing key sizes would reduce much of that benefit, making alternative algorithms more attractive.
Minimal Human Guidance Produced Maximum Results
AI Worked Almost Independently for 60 Hours
Perhaps even more impressive than the mathematical breakthrough itself was the research process.
Claude Mythos operated within an autonomous agent framework for roughly 60 hours.
Only occasional guidance came from a single Anthropic researcher who was not an expert in lattice cryptography.
The AI first reviewed existing academic literature.
It then performed extensive mathematical reasoning.
Afterward, it designed computational experiments to validate its own hypothesis.
Finally, Claude constructed an end-to-end verification pipeline to confirm its findings before presenting them to researchers.
Multiple AI Agents Debated Their Own Research
Independent Reasoning Increased Confidence
Anthropic revealed an especially fascinating aspect of the experiment.
Two separate AI worker agents investigated the same mathematical idea independently.
One concluded the approach was impossible.
The other discovered a successful exploitation technique.
The agents exchanged evidence, challenged one
This resembles scientific peer review, except the reviewers were AI systems.
The computational cost for this experiment reached approximately $100,000 in API usage.
Claude Also Improved AES Cryptanalysis
Reduced-Round AES Research Advances
Claude’s second major contribution focused on AES.
AES protects enormous portions of
Fortunately, Claude did not break the full AES-128 algorithm used globally.
Instead, the AI developed a new fingerprinting technique called the Möbius Bridge for attacking a reduced seven-round version of AES.
The method eliminated one guessing step required during meet-in-the-middle cryptanalysis while introducing multiple optimization techniques to compensate for additional computational complexity.
The resulting attack became approximately 200 to 800 times faster than previous approaches.
Although academically significant, it does not threaten production encryption.
AI Needed Days, Humans Needed Weeks
Validation Became the Real Challenge
Claude spent approximately three days generating hundreds of millions of tokens while exploring cryptographic ideas.
Human researchers, however, spent nearly an entire month verifying the correctness of those results.
Ironically, generating new scientific knowledge required less time than confirming that the AI was actually correct.
This changing balance may become increasingly common across scientific disciplines.
Additional Cryptographic Discoveries
Success Extended Beyond Two Algorithms
Anthropic also disclosed several additional research successes.
Claude developed a practical attack against 13-round LEA, dramatically outperforming previously published methods.
The model also achieved preliminary improvements involving:
Serpent-128
Salsa20
Poseidon hash function
SHA-1 analysis
Each represents incremental scientific progress rather than immediate security risks.
Anthropic responsibly disclosed all relevant findings to affected researchers before making them public.
Responsible Disclosure Protected the Community
Researchers Coordinated Before Publication
Anthropic followed established cybersecurity disclosure practices.
The company privately informed
The findings were also shared with the NIST cryptography mailing list during coordinated disclosure.
This approach ensured researchers had time to evaluate the implications before public discussion intensified.
The Bigger Picture
Artificial Intelligence Has Entered Scientific Discovery
Perhaps the most important conclusion is not any individual attack.
It is the speed of AI progress.
Only one year ago, language models struggled with elementary cryptanalysis exercises.
Today, AI systems can independently generate original research that challenges algorithms examined by human experts for years.
That pace of improvement suggests AI may become a permanent research partner across mathematics, cybersecurity, chemistry, physics, and engineering.
Deep Analysis
Understanding Modern Cryptanalysis Through Practical Commands
The discoveries discussed in this research are theoretical and aimed at strengthening future cryptographic standards. Security researchers often use the following safe tools and commands to study cryptographic implementations in controlled environments.
Check OpenSSL Version
openssl version -a
Generate AES Test Data
openssl rand -hex 32
Encrypt a File Using AES-256
openssl enc -aes-256-cbc -salt -in plaintext.txt -out encrypted.bin
Benchmark AES Performance
openssl speed aes
Verify SHA-256 Hash
sha256sum sample.txt
Display Supported Cipher Suites
openssl ciphers -v
Clone PQC Research Repository
git clone https://github.com/open-quantum-safe/liboqs.git
Build Open Quantum Safe Library
cmake -GNinja . ninja
Run Cryptographic Test Suite
ctest
These commands illustrate how researchers evaluate cryptographic implementations, benchmark algorithms, and experiment with post-quantum libraries in secure laboratory environments. They are intended for educational and defensive research purposes.
What Undercode Say:
AI Has Crossed an Important Scientific Threshold
Claude Mythos is no longer acting as an assistant that accelerates research. It has demonstrated the ability to originate mathematical ideas that experts had not previously identified. That distinction changes how AI should be evaluated in scientific environments.
Validation Is Becoming the New Bottleneck
The most surprising detail is not that AI generated new attacks. It is that human experts spent far more time verifying the discoveries than the AI spent producing them. Future research teams may require entirely new validation workflows.
Cryptography Enters an AI Era
For decades, cryptographic progress relied on relatively small communities of specialized mathematicians. AI may dramatically increase the pace of theoretical exploration, allowing thousands of candidate attacks to be evaluated simultaneously.
Post-Quantum Standards Must Evolve Faster
NIST’s standardization process already involves years of peer review. AI-assisted cryptanalysis could require continuous reevaluation of candidate algorithms throughout their lifecycle rather than only during initial selection.
Agent Collaboration Is Equally Important
The interaction between multiple AI agents demonstrates that disagreement and independent verification can improve research quality. Future AI laboratories may resemble teams of specialists rather than a single large model.
Research Costs Will Continue Falling
Although this experiment reportedly consumed around $100,000 in compute resources, AI infrastructure continues becoming cheaper. Similar discoveries may eventually cost only a fraction of today’s expense.
Security Benefits Outweigh Immediate Risks
None of the attacks threaten production systems. Instead, they help identify weaknesses before standards become widely deployed, strengthening global cybersecurity in the long run.
Academic Publishing May Change Dramatically
Universities and research institutions could soon receive AI-generated papers requiring extensive human validation. Peer review itself may become one of the most valuable scientific skills.
Cybersecurity Teams Need Mathematical Expertise
Organizations have traditionally focused on implementation flaws. Future defensive teams may increasingly need researchers capable of evaluating AI-generated mathematical attacks against cryptographic primitives.
Trust Will Depend on Verification
The future will not belong to whoever generates the most discoveries, but to those capable of proving their correctness efficiently. Independent verification frameworks may become just as important as AI innovation itself.
Prediction
(+1) AI Will Become a Permanent Member of Elite Cryptography Research Teams 🚀
Within the next few years, frontier AI systems are likely to participate directly in the design, analysis, and validation of next-generation cryptographic standards. Rather than replacing human cryptographers, they will dramatically accelerate discovery while experts focus on rigorous verification, formal proofs, and real-world security analysis. Organizations developing post-quantum encryption will increasingly rely on AI to stress-test algorithms long before they reach production, ultimately leading to stronger and more resilient global security standards.
✅ Verified Research Announcement
Anthropic publicly announced that Claude Mythos Preview autonomously discovered new cryptographic research results involving HAWK and reduced-round AES. These findings were documented alongside technical research papers.
✅ No Production Encryption Was Broken
The reported attacks target a post-quantum candidate algorithm and reduced-round cryptographic variants used for academic analysis. There is no evidence that today’s deployed AES-128 or other production encryption systems were compromised.
✅ AI Research Capabilities Are Advancing Rapidly
The broader conclusion that frontier AI is becoming capable of meaningful scientific and cryptographic research is supported by the reported experiments. However, human verification remains essential before any AI-generated discovery can be accepted as accurate or influence future cryptographic standards.
▶️ Related Video (76% Match):
🕵️📝Let’s dive deep and fact‑check.
🎓 Live Courses & Certifications:
Join Undercode Academy for Verified Certifications
🚀 Request a Custom Project:
Secure, high-velocity infrastructure and disruptive technological engineering. Contact our engineering team for high-tier development and proprietary systems:
[email protected]
💎 Smart Architecture | 🛡️ Secure by Design | ⭐ Trusted by Thousands
References:
Reported By: securityaffairs.com
Extra Source Hub (Possible Sources for article):
https://www.reddit.com
Wikipedia
OpenAi & Undercode AI
Image Source:
Unsplash
Undercode AI DI v2
🔐JOIN OUR CYBER WORLD [ CVE News • HackMonitor • UndercodeNews ]
📢 Follow UndercodeNews & Stay Tuned:
𝕏 formerly Twitter 🐦 | @ Threads | 🔗 Linkedin | 🦋BlueSky | 🐘Mastodon | 📺Youtube
