Stanford University
Oct 12, 2026

Stanford cryptographer Dan Boneh says AI can strengthen trust in cryptography

Stanford University cryptographer Dan Boneh said in a post on X that artificial intelligence can, in fact, increase people’s confidence in cryptography. Boneh also said that attempts to break cryptography are the best way to make it stronger. The brief was published by Techub News and cited @a16zcrypto as the source tag. No additional details were provided in the original item beyond Boneh’s remarks on X and the attribution included with the post.

830
Stanford cryptographer Dan Boneh says AI can strengthen trust in cryptography
Starknet
Oct 12, 2026

Starknet says users can upgrade cryptography without changing addresses, after comparing 10 networks’ quantum readiness

Starknet said in a post on X that most blockchains do not have a quantum-safety plan and would have to rely on fund migration to respond to quantum-related risks. The network said its own design allows cryptographic schemes to be upgraded without requiring users to change their addresses, which means funds would not need to be moved during such an upgrade. Starknet described this capability as "cryptographic agility" in the context of preparing for quantum risk. The project also said it has compared the quantum-security preparedness of 10 different networks. The statement was cited by Odaily in a 7x24 news brief.

860
Starknet says users can upgrade cryptography without changing addresses, after comparing 10 networks’ quantum readiness
a16z
Oct 12, 2026

a16z and Stanford cryptography professor say AI math gains have not broken crypto assumptions

OpenAI’s latest progress in mathematical reasoning has sparked concern that AI could undermine the cryptographic systems that support modern software and digital assets. In a discussion hosted by a16z crypto, Stanford University cryptographer Dan Boneh and a16z crypto executives and researchers argued that the recent results do not show any direct advance against core cryptographic assumptions. They said there is no evidence that elliptic-curve cryptography is now weaker than before, and nothing announced this week came close to resolving the P versus NP problem that sits beneath many discussions of computational hardness. The speakers also framed AI as a practical tool for defense. Boneh said models could compress years of cryptographic review by testing new assumptions at much higher speed before systems are deployed. The panel added that some mitigation work already underway for quantum threats may help with parts of the AI risk discussion, though AI could also examine post-quantum systems. A major takeaway for builders was less about collapse and more about process: use AI to stress-test cryptographic designs and wider software stacks before release, and rely on cryptographic agility so systems can switch methods if weaknesses emerge.

1720
a16z and Stanford cryptography professor say AI math gains have not broken crypto assumptions
a16z
Oct 12, 2026

a16z podcast says AI has not broken crypto’s core assumptions, Stanford professor Dan Boneh argues

Andreessen Horowitz’s crypto team used its latest podcast to push back on growing fears that recent AI advances in mathematics could upend modern cryptography. In the discussion, Stanford University professor Dan Boneh said there is no new evidence showing that specific cryptographic assumptions have become weaker, while a16z crypto’s Eddy Lazzarin said nothing suggests elliptic-curve systems are suddenly less secure than before. The panel’s argument centered on a familiar foundation of modern cryptography: many systems rely on the belief that some mathematical problems do not have efficient algorithms at all. If that is true, then a more capable AI model does not automatically change the security picture. The speakers also pointed to a practical upside. AI may compress the long review cycle that cryptographic systems usually require, helping researchers attack designs earlier and more aggressively before deployment. They added that work already underway to prepare for quantum threats has given the industry a head start on some mitigation measures, even if AI could still present broader risks, including to post-quantum schemes. In their view, cryptographic agility and AI-assisted vulnerability detection are now among the most useful defenses.

1120
a16z podcast says AI has not broken crypto’s core assumptions, Stanford professor Dan Boneh argues
Bitcoin
Oct 12, 2026

AI models put Bitcoin signature break risk as high as 76% over 20 years

On Oct. 11, on-chain analyst Willy Woo shared AI-generated estimates on the odds of Bitcoin digital signatures being broken, putting the probability at 1.3% over the next two years. The estimates covered both quantum computing and classical mathematics or AI-assisted cryptanalysis. Researchers then used about four AI models to assess the chances that Bitcoin’s ECDSA and Schnorr signatures could be practically broken over 2, 5, 10, and 20 years. For the 20-year horizon, ChatGPT-6 Astra gave the highest estimate at about 76%, followed by Claude at about 66%, Grok at 60%, and Deepseek at 34%. All four models treated quantum computing as the main long-term threat, while placing the two-year probability in a 0% to 2% range. The models also said there is currently no practical classical cryptanalytic attack against secp256k1, and that stronger AI capability on its own has not yet produced a verified cryptanalytic breakthrough. The report was cited by Bitcoin.com News.

1020
AI models put Bitcoin signature break risk as high as 76% over 20 years
a16z crypto
Oct 12, 2026

a16z crypto panel says recent AI advances have not weakened core cryptographic assumptions

a16z crypto brought together experts including Stanford professor Dan Boneh and a16z crypto partner Eddy Lazzarin to discuss whether fast-moving AI research poses a direct threat to modern cryptography. The view presented in the discussion was cautious rather than alarmist: recent progress in mathematics-related AI work has not produced cryptographic breakthroughs, nor has it shown that the security of any specific cryptographic assumption has deteriorated. The experts said modern cryptography rests on assumptions that certain mathematical problems do not have efficient algorithms. If that assumption holds, stronger AI alone would not be able to discover an algorithm that does not exist. They linked that idea to the still-unsolved P versus NP problem in computer science and said current AI progress remains far from reaching that frontier. At the same time, the panel argued AI could materially shorten cryptographic review cycles. Work that once might have taken a decade to pressure-test could now be challenged by running models for a month in an attempt to break an assumption. They also noted that migration efforts driven by quantum risk have already prepared the ground for part of the AI-related threat model, while cryptographic agility gives systems a way to switch methods if weaknesses emerge. For developers, the most immediate use case is using AI to find more bugs before software goes live.

860
a16z crypto panel says recent AI advances have not weakened core cryptographic assumptions
Policy and Regulation
Oct 10, 2026

AI-driven math advances spark calls for blockchain cryptography recovery plans

A PANews article translated from Dragonfly partner hosseeb argues that the crypto industry should start preparing for a low-probability but catastrophic failure of public-key cryptography if AI produces unexpected mathematical breakthroughs. The piece points to OpenAI’s release of 722 math papers, comments from Scott Aaronson, Matthew Green, Justin Drake and Vitalik Buterin, and says the real concern is not faster quantum computing but a direct break in the assumptions behind systems such as ECDSA. In that scenario, ordinary computers could potentially solve discrete logarithms, making modern public-key cryptography unusable. The article rejects panic and also questions individual “bunker mode” strategies that focus on moving funds to fresh addresses. Its core argument is that blockchains cannot rely on user-by-user self-protection if the cryptographic base layer fails. Instead, major chains should pre-arrange a protocol-level and ecosystem-wide emergency framework, including backup hash-based signature systems, validator-triggered emergency switches, and recovery paths for users who did not migrate in time. The proposal is presented as rough and incomplete, but the author says the industry needs a coordinated plan before any crisis arrives.

960
AI-driven math advances spark calls for blockchain cryptography recovery plans
Policy Regulation
Oct 10, 2026

ChainFeeds Research Roundup: AI Cryptography Risk, Ethereum Glamsterdam, and Q3 Crypto Funding

ChainFeeds’ Oct. 10 research roundup pulled together several major themes shaping crypto markets and infrastructure. One essay argued that rapid AI-driven progress in mathematics has turned cryptographic failure from a remote thought experiment into a governance problem blockchains should prepare for now, including through a prebuilt “Cryptographic Recovery Mode” based on hash signatures. Another focused on Ethereum’s next major upgrade, Glamsterdam, which went live on the Sepolia testnet on Oct. 6 and is designed to let the network safely move toward a 200 million gas ceiling, more than triple the current 60 million target. The report also examined why crypto prices and business funding diverged in the third quarter of 2026: Bitcoin rose 43%, U.S. spot Bitcoin ETFs saw $6.34 billion in net inflows, and sentiment improved, yet venture activity, M&A size, and seed-stage appetite remained weak. Other pieces covered the financialization of AI compute, where hundreds of billions of dollars in GPU assets may eventually be traded and financed through crypto rails, and Starknet’s early discussion of leaving Ethereum to become an independent Layer 1 by 2027 with post-quantum and AI-resistant positioning. Together, the briefing showed a market where infrastructure, security assumptions, and capital formation are all being reworked at the same time.

1090
ChainFeeds Research Roundup: AI Cryptography Risk, Ethereum Glamsterdam, and Q3 Crypto Funding
Total 127 items
Go topage