i · ii Inside and outside the boundary

The threshold scheme defeats one class of risk.

Threshold custody is a precise tool. It removes single points of failure from how a secret is stored — and it does that very well. It does not, by itself, defend against compromise of the devices you use, the people you trust, or the procedures you follow. Both columns below are equally important to understand.

Protected against
Single points of failure
Risks the threshold scheme defeats by design
  • Losing any one shard
  • Theft of any one shard
  • Compromise of any one holder
  • A house fire or local disaster destroying one location
  • One person attempting unilateral recovery
  • Single-vault or single-lawyer custody failure
  • Shards (the company) ceasing operations
  • Any participant being unavailable below the redundancy threshold
Outside the guarantee
Risks the cryptography cannot reach
  • Malware on the splitting or recovery device
  • A spoofed Shards website (phishing)
  • Browser extensions reading input fields
  • Printer memory retaining print jobs
  • Screenshots or cloud-synced clipboards
  • Storing several shards in one location
  • Losing too many shards to meet the threshold
  • Mistyping the original secret before splitting
  • Coordinated coercion of enough holders to reconstruct
The boundary is the math

Information-theoretic security applies to the split itself. Fewer than M shards mathematically reveal nothing about the secret. That is the entire guarantee. Everything outside it — the room you split the secret in, the people who hold the shards, the printer that committed them to paper — is procedural, and procedure can fail.

This is not a flaw in the design. It is the design. Shards solves the cryptographic problem cleanly, on the assumption that the user will solve the procedural problems with care. The best practices section exists for exactly this reason.

iii Information theory, not computational hardness

What "quantum-safe" actually means.

Most cryptography is "secure" in the sense that breaking it would take longer than the age of the universe — provided no faster algorithm is discovered and no fundamentally different machine is built. Shamir's Secret Sharing is secure in a stronger sense: the missing information simply is not present in the available shards, regardless of what computer attempts to recover it.

A polynomial of degree M − 1 has exactly M coefficients. If an attacker has M − 1 shards, they have M − 1 points on the polynomial. Infinitely many polynomials of degree M − 1 pass through any M − 1 points. Each one has a different constant term — that is, a different secret. From the attacker's point of view, every possible secret remains equally likely.

A faster algorithm cannot help, because there is nothing to search. A quantum computer cannot help, because the ambiguity is not encoded in any structure to attack. This is what "information-theoretic" means: the security does not rest on what a future computer can compute, but on what is mathematically absent.

What quantum-safe does not protect

"Quantum-safe" is a property of the threshold split. It is not a property of:

— the device you perform the split on, which may be compromised today by classical malware
— the printer that committed your shares to paper, which has classical firmware
— the secret you started with (e.g., a BIP-39 seed protecting a cryptocurrency address whose elliptic-curve signature scheme is not quantum-safe)
— the holders themselves, whose trustworthiness is unrelated to mathematics

If the underlying secret protects an asset on a system whose signatures are vulnerable to future quantum attack, threshold custody does not change that. It protects the storage and recovery of the secret. It does not modify what the secret unlocks.

iv The honest list

Threats we cannot fix.

A threat model is only useful if it names the failures it does not address. Below is the full list of risks that fall outside Shards' boundary, with our assessment of how serious each is in practice and what you can do about it.

Threat Status What you can do
Compromised splitting device Outside Use a clean device, ideally air-gapped. Verify the build hash. Disconnect from the internet before pasting the secret.
Phishing — spoofed Shards site Outside Verify the URL. Run from source after cloning the public repository. Bookmark the real domain.
Browser extension key-logging Outside Use a fresh browser profile or incognito session with no extensions during the split.
Printer firmware retaining jobs Outside Use a personal printer, not an office or shared one. Print and forget — or write shards by hand.
Cloud-synced clipboard or screenshots Outside Disable clipboard sync. Avoid screenshots of any view that contains a secret or share.
Storing multiple shards in one location Outside Physical separation is the entire point. One file, one safe, one envelope per shard. Different rooms, ideally different buildings.
Losing too many shards to reconstruct Outside Choose redundancy in your threshold. 3-of-5 tolerates two losses. 5-of-7 tolerates two losses with more participants involved.
Mistyping the secret before splitting Outside Always test with a dummy secret first. Verify the round-trip reconstruction matches before splitting the real secret.
Coordinated coercion of M holders Outside Choose holders across legal jurisdictions. Inform holders that no individual coercion can extract the secret. M-of-N raises the cost of coercion, but does not eliminate it.
Single lost shard Protected Nothing required — the threshold scheme tolerates loss of N − M shards by design.
Single compromised holder Protected Nothing required — a single shard reveals nothing about the secret.
Shards (the company) shutting down Protected Nothing required — the open-source library and standalone HTML let you recover with or without us.
The most important thing on this page
A perfect cryptographic primitive applied carelessly is not better than an imperfect one applied with discipline.

If you split a high-value secret on a laptop with one malicious browser extension, then print the shares on an office printer with a network spool, then photograph them and email them to yourself "for backup" — the threshold scheme has done its job, and you have still failed.

The list above is not theoretical. Every one of these mistakes has been made by sophisticated users. Read it, then read it again before you split anything that matters.

v Closing the procedural gaps

Eight rules for real deployments.

The cryptography only protects what the procedure preserves. These eight rules close the gaps the math cannot. They appear in every kit we ship and on the security page — but they belong in the threat model too, because the procedure is the threat surface.

i
Test first with a dummy secret
Verify the round-trip reconstruction before splitting anything you cannot afford to lose.
ii
Use offline mode for high-value secrets
Disconnect from the internet after the page has loaded. Or clone the repo and run from source.
iii
Store shards in separate physical locations
The threshold protection assumes physical separation. Same building, same risk.
iv
Never photograph all shards together
One image of five shares defeats five-point distribution. Same for screenshots and cloud uploads.
v
Do not email shards to yourself
Email is one location, no matter how many addresses you copy. Cloud-backed mail is also a server you don't control.
vi
Use a trusted printer or write by hand
Office printers commonly retain print jobs in memory. Personal printers are safer. Pen and paper are safest.
vii
Keep recovery instructions with each shard
Holders should know how to use what they hold. The standalone HTML and the printed manual are designed for this.
viii
Tell trusted holders what their shard is for
Silent shards are often discarded as junk. Discreet is fine; mysterious is dangerous.
Disclosure policy

If you discover a vulnerability in the Shards cryptographic core, please disclose it to us privately at [email protected]. Our policy follows a standard 90-day responsible disclosure window: we acknowledge within 48 hours, work toward a fix within 90 days, and credit reporters who request it.

The full disclosure policy will be published as SECURITY.md in the public repository when it goes live.