Mark RUSTSEC-2026-0273 as patched in manzana 0.3.0

[noahgift]

manzana 0.3.0 is published and fixes this advisory. I am the maintainer of manzana.

patched = [">= 0.3.0"]

Fixed by removing the cryptography rather than implementing it. The secure_enclave module and the Security-framework FFI are deleted. 0.3.0 exports no key, signature or signer type, makes no Security framework call, and declares no cryptography dependency. Users wanting Secure Enclave or Keychain access are pointed at security-framework.

Implementing it properly needs an Apple Developer account and a code-signed binary with a Team ID; without one SecKeyCreateRandomKey returns errSecInteractionNotAllowed. Publishing cryptography that cannot be executed end-to-end is how this happened in the first place.

0.1.0 and 0.2.0 remain yanked.

Also dropping arch = ["aarch64"]

The affected code was not aarch64-specific, and the narrower constraint under-reported who was exposed.

SecureEnclaveSigner::is_available() was #[cfg(target_os = "macos")] true — a compile-time constant with no runtime probe — so it returned true on x86_64 macOS hosts with no Secure Enclave at all, and sign()/verify() "succeeded" there exactly as they did on Apple Silicon. Intel Mac users were equally affected and, with arch = ["aarch64"] set, were not warned by cargo audit.

os = ["macos"] is kept.

Note

Auditing for the reported defect turned up the same fabricate-a-plausible-value pattern outside the module this advisory names — neural_engine::infer() returning Tensor::zeros(input.shape), capabilities() returning one chip's published figures on every chip, metal::dispatch() returning Ok(()) having dispatched nothing. All removed in the same release; every operation that cannot reach hardware now returns Err(Error::Unimplemented). Mentioned only for context — none of it changes the scope of this advisory, which is about the Secure Enclave signer.

Repository: https://github.com/paiml/manzana
Report: paiml/manzana#3

[djc]

Thanks for the updates!

[djc]

I don't think any of the added advisory contents are useful here. You can explain this stuff in the 0.3.0 release notes.

[noahgift]

@djc friendly ping on this one when you have a moment. It marks RUSTSEC-2026-0273 as patched in manzana 0.3.0.

Status since the advisory was assigned:

• manzana 0.3.0 and 0.3.1 are published on crates.io; 0.1.0 and 0.2.0 are yanked.
• The fix (fix(security): remove all fabricated results; delete the cryptography surface (RUSTSEC-2026-0273) paiml/manzana#4) deletes the secure_enclave module and the rest of the cryptography surface entirely, rather than replacing the stubs, so none of the functions listed under [affected.functions] exist in the patched versions.
• The lint check on this PR is green.

Happy to adjust the entry if you would prefer a different shape (for example unaffected vs patched).

[djc]

I already added review feedback:

#3193 (comment)

[noahgift]

Done — dropped the added prose; the diff is now just the two metadata lines (patched = [">= 0.3.0"], and removing arch = ["aarch64"]). The rationale lives in the commit message and the 0.3.0 release notes instead.

The arch removal I kept, since it changes who gets warned rather than just describing the fix: is_available() was a compile-time #[cfg(target_os = "macos")] true with no runtime probe, so Intel Macs with no Secure Enclave were equally affected but excluded by the aarch64 constraint. Happy to drop that hunk too if you would rather keep this PR to the patched bump alone.