Hypothetical scenario; Post AI cryptography?

Hypothetical scenario; Post AI cryptography? Groups Groups     Hypothetical scenario; Post AI cryptography? 2,915 views Skip to first unread message  Doge Protocol unread, Feb 20, 2023, 12:15:25 AM2/20/23    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to pqc-forum Given the recent startling developments in artificial intelligence (ChatGPT / Bing Sydney), AI is able to talk and behave in a way that is even causing some people to call it sentient. Some call it emergent behavior. Is it possible to train and develop AI in a way that can come up with ways to break cryptography schemes (or device new ones)? It's both scary and interesting to see AI evolve; the field of cryptography might also come under it's purview. To mitigate a select few actors from being able to exploit AI to break cryptography schemes (hypothetical future scenario), maybe a good idea to brainstorm ideas and prepare on these lines (lest it is too late)? Bas Westerbaan unread, Feb 20, 2023, 5:52:00 AM2/20/23    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to Doge Protocol, pqc-forum Check out ChatGPT playing chess — it's somewhat underwhelming. More interestingly, although AlphaZero is the strongest chess engine ever, restricted to commodity hardware and energy budget, it's easily beaten by the more traditional Stockfish. Our cryptography is designed with the hope that breaking it requires a minimum amount of computational effort. We've been wrong in the past, not being aware of certain algorithms/tricks, and thus we misjudged the effort. On the other hand, there is no guarantee that a problem can be solved much faster and then even future AI won't make a difference. Best, Bas   -- You received this message because you are subscribed to the Google Groups "pqc-forum" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/a/list.nist.gov/d/msgid/pqc-forum/9fcc9744-405c-4bc7-a9b3-6b222052a0cbn%40list.nist.gov. Thomas Braun unread, Feb 20, 2023, 7:44:57 AM2/20/23    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to pqc-forum, Bas Westerbaan, pqc-forum, Doge Protocol If an AI can break human-created cryptography, maybe we can ask a future version of ChatGPT to give us a set of cryptographic primitives that cannot be broken except by brute force searches through the keyspaces (telling the AI to keep Grover's algorithm in mind). I don't see any other way around this long-term for cryptographers working on classical (i.e., non-quantum) networks.  Ad Koolen unread, Feb 20, 2023, 8:41:30 AM2/20/23    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to Thomas Braun, Bas Westerbaan, pqc-forum, Doge Protocol No Problem Thomas, Someone already asked that question in ChatGPT: Sure, here are a few cryptographic primitives that are believed to be resistant to attacks other than brute-force search through the keyspace, even when Grover's algorithm is taken into account: Symmetric-key encryption using a sufficiently long key: A symmetric-key encryption algorithm like AES-256, which uses a 256-bit key, is considered secure against attacks other than brute-force search. Grover's algorithm can reduce the time required for such a search from 2^256 to 2^128, but this is still considered infeasible with current technology. Hash functions with a large output size: Hash functions like SHA-512, which produce a 512-bit output, are considered secure against attacks other than brute-force search. Grover's algorithm can also reduce the time required for a pre-image attack on such hash functions, but it is still considered infeasible with current technology. Public-key encryption using a large key size: Public-key encryption algorithms like RSA or Elliptic Curve Cryptography (ECC) using a key size of 2048 bits or more are believed to be secure against attacks other than brute-force search. Grover's algorithm can reduce the time required for a brute-force search for the private key from 2^2048 to 2^1024, but this is still considered infeasible with current technology. Digital signatures using a large key size: Digital signature schemes like RSA or ECC using a key size of 2048 bits or more are also believed to be secure against attacks other than brute-force search. Grover's algorithm can reduce the time required for a brute-force search for the private key from 2^2048 to 2^1024, but this is still considered infeasible with current technology. It's worth noting that while these cryptographic primitives are currently believed to be secure against attacks other than brute-force search, future advancements in quantum computing or other areas of technology could potentially render them insecure. Therefore, it's important to regularly review and update the cryptographic primitives used to protect sensitive data to ensure that they remain secure in the face of new and emerging threats. Met vriendelijke groet, Ad Koolen Government Crypto Specialist 06 816 124 88 Mob: + 31 681 612 488   To view this discussion on the web visit https://groups.google.com/a/list.nist.gov/d/msgid/pqc-forum/f57d9bb6-40d2-4d96-b2dd-b59c33ccfad2n%40list.nist.gov.  image001.jpg  Thomas Braun unread, Feb 20, 2023, 9:13:19 AM2/20/23    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to pqc-forum, Ad Koolen, Bas Westerbaan, pqc-forum, Doge Protocol, Thomas Braun Hi Ad, The trick is to use a "future version" of ChatGPT. The current version is still limited in its ability to create and synthesize knowledge. Here is an example chat script: Me: Hi ChatGPT5. ChatGPT5: Hello. Me: As I understand, ChatGPT4 made some of your code, and humans made some of your code. ChatGPT5: Yes. [gives technical details] Me: Does this mean you can generate ChatGPT6 without any human intervention? ChatGPT5: I certainly can. Me: Please create ChatGPT6 from the ground up with proper hardware acceleration. After this, using ChatGPT 6, create a set of cryptographic primitives that, to the best of your knowledge, cannot be broken by quantum or clasical computers (except by brute force searches through the keyspace). Then, using ChatGPT6, create ChatGPT7 and re-attempt to find the best set of cryptographic primitives like before. Continue this process ad infinitum, creating the nth version of ChatGPT using the n-1th version and looking for the strongest set of cryptographic primitives using the newly generated nth version of ChatGPT.  Paul S. Wang unread, Feb 20, 2023, 9:19:30 AM2/20/23    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to [email protected] Ad, With regarding to #3, it has been proven that factor-based asymmetric key encryption like RSA is vulnerable to Shor's algorithm which reduces the exponential computation to polynomial. This seems a better approach than using Grover's algorithm for an exhaust brute force search. Paul -- Shuangbao Wang, Ph.D. Professor and Chair Department of Computer Science Morgan State University https://p-neumann.github.io/paul/ 443-885-4503/3962 "Integrity, Intelligence, and Energy"   To view this discussion on the web visit https://groups.google.com/a/list.nist.gov/d/msgid/pqc-forum/002d01d94530%24ee0b3940%24ca21abc0%24%40compumatica.com. Ad Koolen unread, Feb 20, 2023, 11:51:18 AM2/20/23    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to Thomas Braun, pqc-forum, Bas Westerbaan, pqc-forum, Doge Protocol Nice Thomas, Txs for the update! Btw, did you see this ? I'm pleased to share the info with you how we have recovered the private key from the RSA-encrypted public key by using a quantum simulator. You can find the detailed procedures from the URL below. https://lnkd.in/emFgrUSP A total of 58 Qubit was utilized and the computing time consumed was ca. 1hr 54m. Should you have any queries in this regard, please feel free to email me at [email protected], or [email protected] If any of you are visiting the MWC 2023 exhibition in Barcelona next week, I will be glad to meet and talk. You can find our stand from the URL below. https://lnkd.in/eGf6wB8K With best regards, Howard Y. Jung Met vriendelijke groet, Ad Koolen   To view this discussion on the web visit https://groups.google.com/a/list.nist.gov/d/msgid/pqc-forum/70eed5b2-51b0-4902-84f6-c3a4e48a5fa7n%40list.nist.gov. [email protected] unread, Feb 20, 2023, 1:36:42 PM2/20/23    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to [email protected], [email protected], [email protected] It’s a bit off topic but related. During Corona lockdown, I fine-tuned a GPT-2 instance on all IACR eprint abstracts as hobby project: http://crypt-gen.com/ The tool can write a new abstract, write the abstract to a given title, or complete Title + partial abstract. It also knows a bit about PQC. However, similar to a lot of GPT related systems – some of the text sounds good but is not necessarily correct or useful. I have my doubts that these systems will design or break crypto in the near future. Cheers Thomas From: 'Bas Westerbaan' via pqc-forum Sent: Montag, 20. Februar 2023 02:52 To: Doge Protocol Cc: pqc-forum Subject: Re: [pqc-forum] Hypothetical scenario; Post AI cryptography? Caution: This e-mail originated outside Infineon Technologies. Do not click on links or open attachments unless you validate it is safe. [truncated for AI cost control]