Mistakes in cryptographic tooling are making headlines!
First, it was using the number 4 as a private key. Now it's using AES keys with only 9 bits of randomness (they should have 128, 192, or 256).
So, it feels like a good time to share something we’ve been working on for a ryptography will play a critical role in securing the world in which AI agents reign supreme. High-assurance building blocks like encryption, cryptographic signatures, and key derivation provide guarantees that other security controls can’t.
But that’s only true if you use them correctly.
That’s why we built Vitamin C: Misuse-resistant building blocks for cryptography.
Strong cryptography depends on more than choosing a secure algorithm. Keys need sufficient randomness, secrets need careful handling, and encrypted data needs to be bound to the context in which it is used. Getting these details wrong can undermine the guarantees of the underlying primitives.
Vitamin C is a suite of Rust crates that makes these details easier to get right - for you or your agent. It provides components for protecting secret values, generating cryptographic randomness, handling keys and passwords, encrypting structured values, and unambiguously encoding cryptographic context. It builds on libraries such as Subtle, Zeroize, and RustCrypto, with types and APIs designed to prevent misuse. The goal is to make invalid states and configurations fail to compile wherever possible.
Experts on our team have thoroughly audited each component and comprehensively tested it with property and mutation testing, fuzzing, and formal verification where possible.
Vitamin C is fully open-source and powers much of what we’re building at CipherStash. We’re targeting a 1.0 release by the end of the year and welcome feedback and contributions.
First, it was using the number 4 as a private key. Now it's using AES keys with only 9 bits of randomness (they should have 128, 192, or 256).
So, it feels like a good time to share something we’ve been working on for a ryptography will play a critical role in securing the world in which AI agents reign supreme. High-assurance building blocks like encryption, cryptographic signatures, and key derivation provide guarantees that other security controls can’t.
But that’s only true if you use them correctly.
That’s why we built Vitamin C: Misuse-resistant building blocks for cryptography.
Strong cryptography depends on more than choosing a secure algorithm. Keys need sufficient randomness, secrets need careful handling, and encrypted data needs to be bound to the context in which it is used. Getting these details wrong can undermine the guarantees of the underlying primitives.
Vitamin C is a suite of Rust crates that makes these details easier to get right - for you or your agent. It provides components for protecting secret values, generating cryptographic randomness, handling keys and passwords, encrypting structured values, and unambiguously encoding cryptographic context. It builds on libraries such as Subtle, Zeroize, and RustCrypto, with types and APIs designed to prevent misuse. The goal is to make invalid states and configurations fail to compile wherever possible.
Experts on our team have thoroughly audited each component and comprehensively tested it with property and mutation testing, fuzzing, and formal verification where possible.
Vitamin C is fully open-source and powers much of what we’re building at CipherStash. We’re targeting a 1.0 release by the end of the year and welcome feedback and contributions.
2
0
3
199
Something weird is happening in tech right now that isn’t getting as much airtime as I expected.
Big tech is acquiring a lot of open-source startups.
Vercel bought BetterAuth
Supabase bought Turso
Cloudflare bought Deno (and killed most of it!)
And many other curious purchases. All for startups with next-to-no revenue and for undisclosed sums.
I suspect the valuation secrecy is less about how desperate the seller was and more about not revealing how motivated the buyer was.
But what’s really interesting is the trend:
Strategic infrastructure is worth more right now than standalone revenue.
Sure, developer adoption is a distribution advantage.
Bun has almost 100k stars on GitHub. BetterAuth 30k. But that’s not the main driver.
AI makes developer infrastructure strategically critical. And almost all of the really important stuff is open source.
Anthropic didn't need Bun to generate significant subscription revenue. It needed Bun to help Claude Code run faster and more reliably. The Bun team described it: if Bun breaks, Claude Code breaks.
This is about acquiring runtimes, build systems, development tools, and identity infrastructure because these are the essential components of agent-driven software development.
The deals involve projects with significant technical credibility, adoption, or exceptionally specialized teams. Because the infrastructure layer is the hardest part to build right!
Even for frontier models, it's often cheaper to buy these capabilities than to build them.
It's also hard to trust infra if it's proprietary. So, to be a force in tech in years to come, it's all open-source, baby!
Big tech is acquiring a lot of open-source startups.
Vercel bought BetterAuth
Supabase bought Turso
Cloudflare bought Deno (and killed most of it!)
And many other curious purchases. All for startups with next-to-no revenue and for undisclosed sums.
I suspect the valuation secrecy is less about how desperate the seller was and more about not revealing how motivated the buyer was.
But what’s really interesting is the trend:
Strategic infrastructure is worth more right now than standalone revenue.
Sure, developer adoption is a distribution advantage.
Bun has almost 100k stars on GitHub. BetterAuth 30k. But that’s not the main driver.
AI makes developer infrastructure strategically critical. And almost all of the really important stuff is open source.
Anthropic didn't need Bun to generate significant subscription revenue. It needed Bun to help Claude Code run faster and more reliably. The Bun team described it: if Bun breaks, Claude Code breaks.
This is about acquiring runtimes, build systems, development tools, and identity infrastructure because these are the essential components of agent-driven software development.
The deals involve projects with significant technical credibility, adoption, or exceptionally specialized teams. Because the infrastructure layer is the hardest part to build right!
Even for frontier models, it's often cheaper to buy these capabilities than to build them.
It's also hard to trust infra if it's proprietary. So, to be a force in tech in years to come, it's all open-source, baby!
9
0
6
1,590
What an epic event! @supabase select was so much fun.
Still so honoured to be partnered with this incredible company.
Still so honoured to be partnered with this incredible company.
1
0
19
707
Anthropic just used Claude to break RSA-896 encryption.
It was only 16-days after RSA-260 was broken and RSA-1024 is the next target.
Should we be concerned?
No. In fact, I think it’s a good thing.
The breakthrough here was to use GPUs instead of CPUs to find prime factors in time to just 10 days instead of literally decades.
Claude didn’t find a new algorithm - it wrote the code to make GNFS (General Number Field Sieves) work in parallel. Seriously impressive but not for the reason that headlines will have you believe.
RSA-896 is old and hasn’t been a standard since 2007.
If you still use RSA, you likely use 2048 or 4096-bit keys.
Are they at risk?
At face value, it seems like it’s just a matter of time before RSA of any size will be easily factored. But the jump from 1024 to 2048 is astronomical compared to the one from 260 to 896.
Assuming the GPU approach scales, factoring RSA-2048 would take ~510 million years on the same 2048 GPUs used to break RSA-896. RSA-4096 would take 430 quintillion years - or 31 billion times the current age of the universe!
Larger RSA keys are perfectly safe from this kind of attack (though the jury is still out on the practicality of quantum attacks any time soon).
So, why do I think this is a good thing?
It reminds us that most cryptography is about computational advantage. And AI driven results like this push technology providers to ensure they use the current best practices.
There’s no hiding anymore - if there’s a weakness, AI will find it, forcing the adoption of better tech. And that’s a good thing!
It was only 16-days after RSA-260 was broken and RSA-1024 is the next target.
Should we be concerned?
No. In fact, I think it’s a good thing.
The breakthrough here was to use GPUs instead of CPUs to find prime factors in time to just 10 days instead of literally decades.
Claude didn’t find a new algorithm - it wrote the code to make GNFS (General Number Field Sieves) work in parallel. Seriously impressive but not for the reason that headlines will have you believe.
RSA-896 is old and hasn’t been a standard since 2007.
If you still use RSA, you likely use 2048 or 4096-bit keys.
Are they at risk?
At face value, it seems like it’s just a matter of time before RSA of any size will be easily factored. But the jump from 1024 to 2048 is astronomical compared to the one from 260 to 896.
Assuming the GPU approach scales, factoring RSA-2048 would take ~510 million years on the same 2048 GPUs used to break RSA-896. RSA-4096 would take 430 quintillion years - or 31 billion times the current age of the universe!
Larger RSA keys are perfectly safe from this kind of attack (though the jury is still out on the practicality of quantum attacks any time soon).
So, why do I think this is a good thing?
It reminds us that most cryptography is about computational advantage. And AI driven results like this push technology providers to ensure they use the current best practices.
There’s no hiding anymore - if there’s a weakness, AI will find it, forcing the adoption of better tech. And that’s a good thing!
2
0
3
367
Mutant tooling is available for most languages. Check out Stryker (JS, Scala, C#), Mutmut (Python) or Gremlins (Go). Leave a comment if you know of others!
Mutation testing is one tool in the kit. Follow me for more.
Mutation testing is one tool in the kit. Follow me for more.
1
0
0
45
*Probably. A failed mutation test doesn’t guarantee that a test is bad but it is almost always a smell!
0
0
0
37
Dan Draper retweeted
Prisma 8 just hit release candidate.
And it has searchable encryption built in thanks to CipherStash.
We’ve worked with the Prisma team to give you:
🔎 Encrypted search
🔽 Encrypted Sorting and range queries
📃 Searchable encrypted JSON
All natively in Prisma 8’s shiny new s support encrypted data
✅ Prisma schema works with encrypted data types
✅ Prisma queries work with encrypted queries
Plus, its fully supported in Prisma Compute and Prisma Postgres.
Data Level Access Control just got even easier. Full announcement in the comments.
And it has searchable encryption built in thanks to CipherStash.
We’ve worked with the Prisma team to give you:
🔎 Encrypted search
🔽 Encrypted Sorting and range queries
📃 Searchable encrypted JSON
All natively in Prisma 8’s shiny new s support encrypted data
✅ Prisma schema works with encrypted data types
✅ Prisma queries work with encrypted queries
Plus, its fully supported in Prisma Compute and Prisma Postgres.
Data Level Access Control just got even easier. Full announcement in the comments.
2
1
11
11,881
Dan Draper retweeted
Today we're launching the CipherStash + Supabase integration.
Developers can now build privacy-first PostgreSQL applications without giving up the features they expect.
What you get:
* Search encrypted data (equality, range, ordering & full-text)
* Encrypt sensitive fields befogres
* Keep keys under your control with ZeroKMS
* Add data-level access control tied to identity
* Works natively with Supabase and TypeScript
This is about making strong security the default—not something that slows developers down.
Learn more:
cipherstash.com/supabase
Docs:
cipherstash.com/docs/integrati…
Thanks to the @supabase team for partnering with us. Looking forward to seeing what everyone builds next.
Developers can now build privacy-first PostgreSQL applications without giving up the features they expect.
What you get:
* Search encrypted data (equality, range, ordering & full-text)
* Encrypt sensitive fields befogres
* Keep keys under your control with ZeroKMS
* Add data-level access control tied to identity
* Works natively with Supabase and TypeScript
This is about making strong security the default—not something that slows developers down.
Learn more:
cipherstash.com/supabase
Docs:
cipherstash.com/docs/integrati…
Thanks to the @supabase team for partnering with us. Looking forward to seeing what everyone builds next.
0
1
3
158
This is the one I've been waiting to share.
Searchable encryption on @supabase is live! Encrypt your columns and keep querying them. WHERE, LIKE, ORDER BY, JOIN, GROUP BY, even JSON. Basically, zero code change and < 10ms overhead for most queries.
Searchable encryption on @supabase is live! Encrypt your columns and keep querying them. WHERE, LIKE, ORDER BY, JOIN, GROUP BY, even JSON. Basically, zero code change and < 10ms overhead for most queries.
CipherStash brings searchable field-level encryption to Supabase.
→ Encrypt sensitive fields at the application layer, no schema changes
→ Search, sort, and join over encrypted data without decrypting it
→ Keys stay under your control with ZeroKMS
One command to set up: `npx Learn more → supabase.com/blog/searchabl…
2
0
1
189