Connect with us

NEWS

eMudhra’s Python Script Turned CA Keys Into the Number 4

eMudhra’s incident report says a one-off Python wrap script, FIPS mode off, truncated eight CA keys to one byte and forced eight more RSA CAs down.

Published

on

eMudhra’s full incident report says a custom Python wrap script truncated eight CA private keys to a single byte. The Bengaluru company, a publicly trusted certificate authority, filed the account on 10 November 2026, almost a month after a researcher showed that those keys could be recovered from the public certificates. One of them was the integer 4.

The hardware security modules used in the ceremony produced usable keys. The approved transfer script that ran beside them is what the company now says it certified.

A Python Script Took One Byte and Certified It

Naveen Kumar ML, the eMudhra engineer assigned on Mozilla’s CA-compliance bug, posted the full incident report on Mozilla’s tracker as comment 82. On 30 April 2025 the PKI operations team generated 16 issuing CA key pairs in the B and C series, eight ECDSA P-384 and eight RSA-4096, on Utimaco modules validated to FIPS 140-2 Level 3. The wrap-and-unwrap step that followed was an approved deviation from the usual ceremony script, written for a one-off move between vendors, and it was the first time that team had used it.

After each wrapped key came back, the script read the unwrapped private value out of the module into host memory and built a new key object from that copy. For the eight ECDSA objects, one line in that read-back kept only the leading byte. The re-created object held a scalar below 256. That object, not the key generated on the module, was used to build the certificate signing request and was certified. The eight values that fell out are 4, 56, 66, 115, 155, 156, 230 and 237.

THE EIGHT CERTIFIED SCALARS

Issuing CAUsePrivate value
EM DV TLS CA – G3BTLS156
EM OV TLS CA – G3BTLS155
EM EV TLS CA – G3BTLS115
EM DV TLS CA – G3CTLS230
EM OV TLS CA – G3CTLS4
EM EV TLS CA – G3CTLS56
EM OV CS CA – G3BCode signing66
EM OV CS CA – G3CCode signing237

A P-384 private key is supposed to be a random integer about 384 bits long. Recovering one from its public point is not a job a laptop finishes. Recovering a key below 256 is a loop of at most 255 guesses. Security researcher Scott Helme recovered all eight from the published certificates in under a second, including the intermediate whose private key was 4.

Wayne, the researcher who opened the incident, found the break at 16:27 UTC on 17 October 2026 while scanning a Common CA Database test corpus. At 18:05 he mailed proof to Apple, Chrome, Microsoft, Mozilla and eMudhra, signing the line “We have a problem here – Wayne” with each key. Mozilla’s Dennis Jackson recovered the same keys independently at 21:48 UTC, with no hint of Wayne’s method, and confirmed that the original generation was flawed rather than the signatures on the certificates.

Why the Keys Were Made Before the Hardware Arrived

The B and C series were meant to run from eMudhra’s US and EU sites, on hardware from a second vendor, Crypto4A. US modules ordered in April 2025 were delayed. Root-store rules taking effect on 15 June 2025 required that subordinate CAs disclosed on or after that date be dedicated to TLS server authentication, and the company also wanted Sectigo cross-certificates in place before that date. So the keys were made on the Utimaco boxes already in a WebTrust-certified hall, with a plan to wrap them for later import.

Cross-vendor wrapping is awkward by design. The team wrote a host-side Python script to bridge it and ran the source module with FIPS-approved mode disabled for that step, recorded as an approved deviation. The 30 April round trip was billed as a rehearsal of the later move. In July the C-series keys did go to the destination site, but the PKCS#11 interface on the new modules could not complete the import, and a custom import was put in place. The defective read-back was not used again in July. The objects that moved were the ones the April ceremony had already re-created, which means the destination modules received the truncated ECDSA keys.

Ian Carroll, a security researcher, put the operational point without dressing it up: eMudhra “disabled FIPS on the HSM, exported the private key into Python on a computer, and then it got truncated when they re-imported it.” An HSM only binds a key that never leaves it in readable form. Once the ceremony asked the module to hand the secret to a Python process, the box was enforcing a policy, not a physical limit.

The same day’s ceremony also produced an A series on a different HSM path. Those keys were later replaced by an A-1 series and are not the keys in this report. Mozilla counted 24 intermediates in the April batch once the A series is included. The wrap step, and the damage, sat on the 16 B and C keys.

A CRL Failure in June Pointed at Two Keys

Nothing in the ceremony tested whether a newly certified CA key was actually strong, or whether the public key in the request matched the key generated on the module. The truncated ECDSA keys still made valid signatures, a valid point on the curve, a CSR whose signature verified, and certificates that linted clean. They were disclosed in the Common CA Database in May 2025. Sectigo cross-signed the eight ECDSA intermediates that month. The hierarchy then passed WebTrust audit cycles in 2025 and 2026. Baseline Requirements do not currently require a strength check on CA keys, and eMudhra did not add one of its own.

There was a warning. In late May 2025 the CRLs for the six G3B and G3C TLS intermediates failed to verify, because the certificates had been issued over the re-created objects while the first CRLs were signed with the originally generated objects. The mismatch was tracked as Mozilla bug 1970728, filed on 5 June 2025, and closed by re-signing with the object that matched each certificate. The report now calls that an early sign of the handling defect. At the time it was treated as a wrong signature, not as proof that two key objects existed for the same CA.

THE 18 MONTHS IN THE OPEN

  1. 30 April 2025: Key ceremony generates the B and C series, eight ECDSA P-384 and eight RSA-4096, then runs the wrap/unwrap script with FIPS-approved mode off.
  2. May 2025: The 16 issuing CAs are disclosed in the Common CA Database. Sectigo issues eight AAA Certificate Services cross-certificates of the ECDSA set.
  3. 29 May to 6 June 2025: CRL signatures for the six G3 TLS intermediates fail, then are re-signed with the matching (weak) objects.
  4. 14 to 15 July 2025: C-series keys move to the destination facility using a custom import. The defective April read-back is not reused.
  5. May 2026: Issuance begins from US issuing CAs. The ECDSA G3B and G3C issuers still issue no end-entity certificates. The RSA G2C TLS and code-signing issuers go into production.
  6. 17 October 2026: Wayne reports the recoverable ECDSA keys at 18:05 UTC. Chrome’s block lands at 19:45. Mozilla finishes its OneCRL update at 22:59.
  7. 18 October 2026: eMudhra revokes all eight ECDSA issuers between 09:24 and 09:38 UTC, reason cACompromise, about 15 hours after Wayne’s mail.
  8. 7 November 2026: The eight RSA G2B and G2C issuers are revoked, reason cessationOfOperation, after a CPS §6.2.6 review.
  9. 10 November 2026: The full incident report is posted as comment 82.

From 30 April 2025 until 17 October 2026 the certified ECDSA keys sat in public, about 18 months on the company’s own count, recoverable by anyone who thought to count up from 1.

Eight RSA CAs Fall for a Memory-Handling Gap

The RSA-4096 objects from the same ceremony were not truncated. Their signatures still verify against the certified moduli. The same host-side pattern still applied: during re-import, private attributes such as CKA_PRIVATE_EXPONENT and CKA_PRIME_1 were handled in the volatile memory of a Python process on the ceremony host. eMudhra says it cannot show that this met CPS §6.2.6, the clause in its own practice statement on moving private keys into or out of a cryptographic module. It therefore treated the RSA material as non-compliant and revoked those eight issuers on 7 November, after saying in October that the RSA path was not hit.

Unlike the ECDSA set, the four RSA G2C issuers had been in production. They started issuing end-entity certificates in May 2026. The last code-signing end-entity from EM OV CS CA – G2C was dated 7 October 2026. The last EM DV TLS CA – G2C end-entity was dated 27 October. The last EM OV TLS CA – G2C end-entity was dated 3 November, the day issuance from those issuers stopped. Subscriber migration from that G2C set is complete, the report says. The eight ECDSA issuers and the four RSA G2B issuers issued no end-entity certificates at all.

FOUR GAPS THE REPORT NAMES

  • Host-side script: A one-off Python wrap/unwrap read the private value into process memory and, for ECDSA, truncated it to the leading byte before the copy was certified.
  • No strength check: The ceremony never tested CA key strength and never compared the certified public key with the key generated on the module.
  • Similar labels: The original object and the re-created object carried almost the same label, so the defective copy could be selected for the request.
  • June treated as a symptom: The 2025 CRL mismatch was fixed by re-signing, without asking why two key handles existed for the same CA.

That is now 16 issuing CAs revoked from one ceremony, plus the eight Sectigo cross-certificates of the ECDSA set, which Sectigo revoked on 19 October. eMudhra says it has no known case of those keys being used to forge a certificate. Chrome’s root program said no TLS certificates chaining to the six weak TLS intermediates were logged in Certificate Transparency before the block, so CT-enforcing clients were not shown a fake site from that set. What cannot be proved is that nobody recovered the keys in the months they were public and used them against software that never checks CT or revocation.

Chrome, Firefox and Windows Pull the Roots

Browsers moved on the day of the report. Chrome completed corroboration and landed a CRLSet update blocking the six TLS keys by subject-public-key hash at 19:45 UTC on 17 October, without a browser restart. Android and Chromebooks received an equivalent update through Play Services around 22:00 UTC on 19 October. Mozilla used OneCRL for blocking the affected certificates in Firefox, finishing at 22:59 UTC, less than three hours after it received the mail at 20:06. On 20 October it shipped Firefox 158.0.1, ESR 115.43.1 and ESR 153.5.1 for users whose settings blocked OneCRL, and it pulled emSign ECC Root CA – G3 from NSS 3.131.1 so other software that borrows Mozilla’s store would drop the root too. An Android follow-up, 158.0.2, went out on 21 October after some profiles skipped the bundled blocklist at startup.

Microsoft blocked the identified intermediates on 20 October, then went wider. An out-of-band Windows update on 26 October, KB5134133, removed seven certificates (six eMudhra G3 roots and an AAA Certificate Services cross-sign) from the Windows trust list for machines that do not take automatic certificate-trust updates. Edge dropped the same seven on 27 October. A 10 November update on the MSRC blog said Microsoft is removing the remaining eMudhra root certificates across its products from 16 November 2026, ten roots in the Trusted Root Program, with Edge, WebView2, Windows, Azure Linux and later.NET and OpenJDK images following on their own clocks.

Chrome had already scheduled removal of emSign ECC Root CA – G3 for the morning of 27 October, Pacific time, in Root Store version 43. On 4 November it added a further cut on emSign Root CA – G1: TLS certificates chaining to that RSA root whose earliest Signed Certificate Timestamp is after 16 November 2026, 23:59:59 UTC, will not be trusted in Chrome 153 and higher. Existing G1 certificates with earlier timestamps stay up in Chrome. eMudhra’s public incident page records the same 16 November SCTNotAfter date for Firefox, and an Apple removal of G1 on 30 November that, unlike Chrome’s cut, drops trust in G1 certificates issued at any time.

The flaw was ours. A script error resulted in issuing CA’s with weak keys.

eMudhra Leadership Team, incident statement, 26 October 2026

The 26 October note also said the affected intermediates were revoked within 14 hours of confirming the issue. The later report times the ECDSA revocations at about 15 hours after Wayne’s 18:05 mail, which is the working figure here. Chairman V. Srinivasan, in the 22 October exit notice, said global SSL/TLS was about 1 percent of eMudhra’s revenue, that no end-entity certificates had been issued from the weak CAs then in view, and that the board was leaving the public TLS business. Digital-signature, eSign, emSigner, CertiNext, emCA, PrivaTrust and SecurePass sit on other infrastructure and, the company says, were not touched.

The Last Hierarchy Stops Issuing on November 16

The G3 ECC distrust still hit a live customer base even though the weak intermediates themselves had issued nothing. eMudhra’s 5 November update put roughly 30,800 certificates under the G3 hierarchies in a single US client ecosystem and said that group had already been given an alternate path through the G1 RSA root. G1 issuance is now only a bridge. Action item 1 in the incident report is a hard stop on all issuance from emSign Root CA – G1 before 16 November 2026, 23:59 UTC, with remaining certificates for a university federation in North America moving to another publicly trusted provider. Chrome and Firefox stop trusting new G1 certificates after that same date. Microsoft’s remaining-root removal begins the same day. Apple’s G1 drop follows on 30 November.

KPMG was named as independent reviewer on 30 October. The report dates that review to 4 December 2026, and a full inventory of CA keys to 30 November. Future ceremonies, eMudhra says, will generate keys in place and will not move them between devices. The script that caused this ran once, on an isolated ceremony host, and was not part of the emCA product. The modules, the company repeats, did what they were asked. The ask was the problem.

Remaining G1 issuance is due to stop on 16 November 2026, the same day Microsoft begins dropping the last 10 eMudhra roots from Windows, Edge and Azure Linux.

Harry is the editor of NEWS ANALYSIS. He writes across the publication's ten desks, with most of his time going to the stories where a number, a filing or a study decides the argument. His working rule is simple: read the source document before writing about it, and tell the reader plainly which parts are established and which are somebody's claim. He is responsible for the standards set out on this site's Editorial Standards and Fact Checking pages, and for correcting the record openly when the publication gets something wrong.

Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Trending