Essay Threat model TNFL

Trust Now, Forge Later — the signature side of the quantum threat

HNDL was the confidentiality story. TNFL — Trust Now, Forge Later — is the integrity story, and it is the harder one. An adversary doesn't need to capture your traffic. They need your public signing certificate. The rest is patience and a sufficiently large quantum computer.

KavachQ's quantum-exposure analysis of Indian BFSI, read through the May 2026 DST/NQM report's TNFL lens: nine systems mapped to crypto-in-use, retention window, and HNDL/TNFL exposure. UPI signing certificates, Aadhaar e-KYC signatures, GST e-invoice DSC chain, NEFT/RTGS message authentication, mobile banking pinning, Income-Tax/MCA DSC, and SWIFT message signing are TNFL-exposed (long-retention signing surfaces on quantum-vulnerable RSA-2048). Internet-banking TLS is HNDL-only (session-scoped, no long-lived signed artefact). ATM PIN-block DUKPT is symmetric and not quantum-exposed (legacy TDES risk only); RSA-based remote key loading is the quantum-relevant part. The footer notes the report's advisory CII Milestone 1 — complete discovery and inventory by 31 December 2027.
Our reading of Indian BFSI signing surfaces through the report's TNFL lens — sorted by retention. The long-retention rows are the TNFL exposure.

The May 2026 NQM Task Force report on India's Quantum-Safe Ecosystem — chaired by Dr. Rajkumar Upadhyay (CEO, C-DOT), with RBI, SEBI, ICICI Bank, CERT-In, IIT Madras and DSCI at the table — formalised an acronym every Indian bank board now needs to know. Not HNDL. That one is old news.

TNFL — Trust Now, Forge Later. The DST framing is summarised on India's PQC roadmap; this post unpacks the mechanic, the exposure inventory, and the December 2027 advisory milestone the report sets for CII sectors, BFSI included.

HNDL — harvest-now, decrypt-later — threatens confidentiality. An adversary records your encrypted traffic today and decrypts it the year a cryptographically-relevant quantum computer (CRQC) arrives. The fix is well understood: encrypt forward today with hybrid PQC for anything with a long secrecy lifetime.

TNFL threatens integrity. An adversary archives your signing certificate today, derives the private key when a CRQC arrives, and then forges signatures the system still treats as authentic. The audit trail will say your bank signed it. The signature will validate. There is no retrospective fix.

The two attacks share an enabling technology (Shor's algorithm on a CRQC) but they hit different security properties and have very different operational tails. HNDL exposure stops growing the day you migrate. TNFL exposure compounds with every long-lived signature you continue to mint on RSA-2048 or ECDSA.

The mechanic, in three frames

This is the part most people miss on the first reading. The attacker doesn't need to capture your transactions. They need one thing — your public signing certificate. And that's already public by design.

Today — archive the certificate

Every CA-issued signing certificate in BFSI is published. UPI signing certificates. Aadhaar e-KYC certificates. GST e-invoice DSCs. SWIFT BIC-linked PKI. Their public keys are the whole point — they let counterparties verify your signatures. An adversary harvests them with a script. RSA-2048. Done.

Q-Day — Shor's derives the private key

When a CRQC of sufficient size exists — mainstream estimates put this anywhere between late this decade and the mid-2030s, with the NQM report noting that most estimates point to a 2028–2032 horizon for practical quantum attacks — Shor's algorithm derives the private key from the public key. This is not "hacking" in the conventional sense. It is mathematics on enough qubits. The cost of the attack drops from infeasible to tractable overnight.

The day after — forged signatures, backdated

With the private key in hand, the adversary can sign anything the original signer was ever authorised to sign. A payment instruction. A KYC response. A SWIFT pacs.008 (formerly MT103). A GST e-invoice. A digitally-signed contract. The forgery is producible in 2031 and dated 2026 if needed. Signed by a key the system still trusts, because nobody has revoked it — and even revocation does not help: the forged document predates any revocation date.

This is the asymmetry that makes TNFL nasty:

You cannot un-sign a forgeable signature after the fact.

Every signature you mint today with quantum-vulnerable cryptography, on an artefact that has to remain trustworthy past Q-Day, is a TNFL liability you are accruing now and cannot recall later.

Where TNFL exposure lives in Indian BFSI

The figure above is our analysis of the signing surfaces the TNFL lens catches first in Indian BFSI. Read the third column — retention — first. Anything where the signed artefact has to remain authoritative for years to decades is a TNFL surface today:

  • UPI signing certificates. RSA-2048, ~7–10 year retention on the underlying transaction records. Every UPI rail message is a candidate TNFL target.
  • Aadhaar e-KYC signatures. RSA-2048, decade-plus retention on the KYC artefacts banks rely on for AML/CFT defensibility.
  • GST e-invoice DSC chain. RSA-2048, ~6 years' retention (72 months from the annual-return due date) under section 36 of the CGST Act.
  • NEFT / RTGS message authentication. RSA-2048, ~10 year retention for settlement reconciliation.
  • Income-Tax & MCA DSCs. RSA-2048, 7+ year retention.
  • SWIFT message signing. RSA-2048, 7–10 year retention under correspondent-banking obligations. Cross-border payments moved from MT to ISO 20022 (MX) messages when coexistence ended in November 2025; the signing exposure is the same either way.
  • Mobile banking certificate pinning. RSA / ECDSA, 1–3 year retention but daily reissuance volume.

Compare to the two systems on the inventory that are not TNFL surfaces — internet-banking TLS and ATM PIN-block DUKPT. Both terminate inside a session. Once the session ends there is no artefact left to forge. Internet-banking TLS is an HNDL surface (the session traffic is recordable and its key exchange is quantum-vulnerable) but not a TNFL surface. ATM PIN-block DUKPT is neither: DUKPT is symmetric key derivation, which Shor's algorithm does not touch, so its only risk is legacy TDES. Remote key loading (RSA) is the quantum-relevant part of the ATM estate. The threat models are not interchangeable, and the distinction matters when you sequence the migration.

The TNFL surface inside an Indian universal bank is large, RSA-2048-dominated, and deeply integrated into rails the bank does not own — UPI, NPCI, GSTN, Aadhaar, SWIFT. Migration is not unilateral.

"The countdown has already begun, and hesitation will be the weakest defence." — Department of Science & Technology, May 2026

Why this is harder than HNDL

HNDL has a clean exit door: encrypt-forward today, the past stays past. TNFL does not. Three reasons:

  • The artefacts are persistent, not ephemeral. A TLS session ends. A signed payment instruction sits in your audit trail for a decade. Whichever signature is on it at issue time is the signature it has forever — unless you re-sign every historical artefact under PQC at some future cut-over, which is operationally implausible.
  • The trust graph is shared. UPI, NPCI, GSTN, UIDAI and SWIFT all sit upstream of your signing posture. Your bank cannot move alone. Migration is a coordinated activity across the entire BFSI rail, with regulators and rail operators setting the cut-over.
  • The fix is dual-signature, not algorithm-swap. A signature mints once and lives forever. Replacing RSA-2048 with ML-DSA on day N protects everything signed on day N+1. Everything signed on day N-1 remains forgeable on Q-Day. The only mitigations for the historical estate are counter-signing under PQC — a second, PQ signature attached to the document — or PQC archive timestamping (RFC 3161 timestamps or RFC 4998 evidence records from a PQ-signed timestamping authority). Most signing infrastructures do not yet support either natively. This is a multi-year change-management problem, not a config flip.

The December 2027 reality check

The Task Force's advisory Milestone 1 for CII sectors, BFSI included, asks for complete discovery and inventory by 31 December 2027 — in practice, a Cryptographic Bill of Materials covering every key, every certificate, every algorithm, every library, in every system the bank operates. The CBOM is the prerequisite, not the migration itself. You cannot migrate what you cannot see.

Nineteen months from this post's publication date.

What does that timeline look like in practice? In a mid-sized PSU bank, cryptographic discovery — the inventory work alone — can take a year or more. The systems are heterogeneous, the documentation is partial, and the certificates are spread across vendors who have not been asked the question before. Migration of the highest-value TNFL surfaces (UPI, Aadhaar, GST, SWIFT) follows a longer tail, gated by NPCI, UIDAI, GSTN and SWIFT respectively.

The honest reading: there is no scenario where Indian BFSI is on track against the December 2027 milestone unless inventory work is already in flight.

What to do this quarter

The first three moves are inexpensive and useful even if your view of Q-Day shifts later:

  1. Inventory the long-lived signing surfaces first. Forget data-in-motion for a moment. Pull the list of every CA-issued signing certificate in production, its algorithm, its expiry, its issuing CA, and the retention window of the artefacts it signs. Sort descending by retention. The top of that list is your TNFL exposure.
  2. Inventory the upstream dependencies. For each long-retention signing surface, identify the external rail operator (NPCI, UIDAI, GSTN, SWIFT, CCA, IDRBT). Your migration will be gated by theirs. Knowing the dependency chain now lets you participate in the operator's consultations rather than react after.
  3. Write PQC into procurement language. Every signing-infrastructure renewal — HSMs, signing services, document- management platforms, CA contracts — should now carry a PQC-capability clause. The cost of inserting the clause today is zero. The cost of swapping a vendor in 2028 because the platform does not support ML-DSA is not.

These are not the migration. They are the prerequisites to having a migration plan worth presenting at the December 2027 milestone. The CBOM is the deliverable. Inventory is what produces it.

Where this leaves you

HNDL was the easier story to tell, and the right story to start with — a cleanly bounded threat with a cleanly bounded fix. TNFL is the harder story and the one most BFSI conversations are not yet having. The NQM Task Force report changes that: TNFL is now in the official vocabulary, the December 2027 inventory milestone is named, and the question of whether the Indian financial system is on track has an answer that does not depend on optimism.

That is the gap KavachQ is being built to close — turning the DST framework into a sequenced, inventoried, audit-ready migration plan against the actual cryptographic estate of an Indian bank. The standards exist. The deadline is on the calendar. The plan still has to be written, per bank, against its real signing surfaces.

Read more
See the platform Run the Maturity Scan Back to the blog