NEWS
eMudhra’s Ceremony Script Cut CA Private Keys to One Byte
eMudhra’s Bugzilla report says a Python wrap test truncated eight P-384 CA keys to one byte, a flaw two WebTrust audits never measured.
eMudhra’s full incident report says a single line in a custom Python script truncated eight CA private keys to one byte during a 30 April 2025 ceremony. The Bengaluru firm posted the account to Mozilla Bugzilla on 10 November 2026, months after browsers had already blocked the keys.
Those eight secrets were ECDSA P-384 values. A researcher who posts as Wayne reported them on 17 October 2026, and one of them was the number 4. The hardware boxes did their job. The wrap test around them did not.
A Wrap Test Turned 48-Byte Secrets Into One Byte
On 30 April 2025, eMudhra generated 24 intermediate CA keys in one sitting: three series (A, B and C), two types (RSA-4096 and P-384), and four jobs (DV, OV, EV, and code signing). Series B was meant for eMudhra BV in the Netherlands. Series C was meant for eMudhra Inc in the United States. Both sat on Utimaco modules certified to FIPS 140-2 Level 3. Series A sat on Entrust modules for India work and was later replaced by an A-1 series.
The B and C keys had to move to machines in other countries, the first time the firm had tried that transfer between hardware security modules in different places. Staff approved a wrap and unwrap round trip, then ran a custom Python key-handling script on the ceremony host to pack each private key into an encrypted blob for transport.
The module created the keys onboard as expected. The script then unwrapped them and read them back. For the eight ECDSA P-384 objects in the G3B and G3C set, that read-back kept only the leading byte. The objects put back on the module were scalars below 256, and those were the keys the root signed and sent onward.
A single line in that read-back truncated the private value to its leading byte, so the re-created object held a scalar below 256.
eMudhra, Full Incident Report, Mozilla Bugzilla 2083449
A normal P-384 private key is a random 48-byte integer. These eight were 47 bytes of zeros and one leftover byte. Because the public point is printed in the certificate, anyone can multiply the curve generator by 1, then 2, then 3, until the point matches. Security researcher Scott Helme recovered all eight on a laptop in less than a second. Mozilla’s Dennis Jackson did the same on the night of the report, without being told Wayne’s method, and noted that a local coding agent recovered them from a one-line prompt.
On 26 October 2026 the leadership team wrote that the flaw was theirs. “A script error resulted in issuing CA’s with weak keys,” the note said. “We want to apologise directly to our customers and to the broader WebPKI community.”
The Eight Numbers Anyone Could Guess
Helme published the private values after walking the public keys. Each number is the entire secret for that issuing CA, and each secret also appears on a Sectigo cross-signed copy. The Data Security Council of India later told operators they could recover a key in under a second on an ordinary laptop, with no special skill.
THE EIGHT P-384 PRIVATE KEYS
| Issuing CA | Purpose | Private value |
|---|---|---|
| EM OV TLS CA – G3C | TLS | 4 |
| EM EV TLS CA – G3C | TLS | 56 |
| EM OV CS CA – G3B | Code signing | 66 |
| EM EV TLS CA – G3B | TLS | 115 |
| EM OV TLS CA – G3B | TLS | 155 |
| EM DV TLS CA – G3B | TLS | 156 |
| EM DV TLS CA – G3C | TLS | 230 |
| EM OV CS CA – G3C | Code signing | 237 |
All eight sit below 256, so the longest search is 237 multiplications. EM OV TLS CA – G3C, the United States OV issuer, is literally 4. Six of the eight could sign website certificates. Two could sign code on Microsoft’s programme. Jackson wrote that the keys were “effectively all-zeroes apart from the last byte.”
Why Two Years of Audits Never Caught a Key of 4
The weak keys still issued certificates as designed. Signatures verified. The public points sat on the P-384 curve. Certificate linters stayed quiet. eMudhra said the hierarchy went through independent audits two years in a row and had been in browser trust lists since mid-2025, covering the 2025 and 2026 WebTrust cycles. The report’s own account is that Baseline Requirements and those audits do not require a test of CA private-key strength, or a check that the certified public key matches the key originally generated on the module.
WHAT THE AUDITS NEVER TESTED
- Key size: No step asked whether a P-384 private scalar was a 48-byte random integer or a number below 256.
- Public match: Staff did not compare the certified public key against the fingerprint recorded at generation.
- Lint and signatures: Valid signatures and clean lint results were treated as proof the ceremony had worked.
- Duplicate labels: The wrap test created a second key pair whose labels looked like the first, and the weak pair was the one sent for signing.
There had already been a warning. In Mozilla bug 1970728, CRLs for the same six TLS issuers carried invalid signatures. Archive copies show those CRLs failing verification from 23 May 2025. eMudhra’s later account is that the lists were first signed with the original strong keys, which did not match the published weak ones, then re-signed with the flawed keys. The bug was closed after corrected lists went out. Nobody then asked why the signing key and the published key were different objects.
The Debian OpenSSL bug of 2008 had the same shape: a programming error collapsed a key space, the signatures still worked, and the certificates stayed trusted until researchers looked at the numbers. eMudhra’s case is narrower and worse in one respect. These were not subscriber keys on random laptops. They were issuing CA keys, created in a witnessed ceremony, then left in public certificate lists from 30 April 2025 until 17 October 2026, a span the company called about 18 months.
Browsers Moved in Hours, and the Root Did Not Survive
Wayne emailed Apple, Chrome, Microsoft, Mozilla, and eMudhra at 18:05 UTC on 17 October 2026, with a signed proof for each key. Chrome’s CRLSet blocked the six TLS public keys by 19:45 UTC, about an hour and 40 minutes later, with no browser restart required. Mozilla received the report at 20:06 UTC, confirmed the proofs, and recovered the keys on its own at 21:48 UTC. Firefox’s OneCRL block was in place by 22:59 UTC. eMudhra acknowledged the report at 21:29 UTC.
FROM THE REPORT TO THE ROOT DISTRUST
- 17 October 2026, 18:05 UTC: Wayne reports eight recoverable keys to Apple, Chrome, Microsoft, Mozilla, and eMudhra.
- 17 October 2026, 19:45 UTC: Chrome CRLSet blocks the six TLS keys.
- 17 October 2026, 22:59 UTC: Firefox OneCRL block completes.
- 18 October 2026, 09:24 to 09:38 UTC: eMudhra revokes the eight intermediates for CA compromise. The company later said that was within 14 hours of confirming the issue.
- 19 October 2026: Apple blocks the certificates. Sectigo revokes the eight cross-signed copies.
- 20 October 2026: Microsoft blocks them. Mozilla removes trust in emSign ECC Root CA – G3 and ships Firefox 158.0.1. Chrome commits to the same root path.
- 27 October 2026: Chrome Root Store Version 43 stops trusting that G3 root on the morning, Pacific time.
Up-to-date browsers were protected the same evening. The harder gap is everything that does not use CRLSet or OneCRL. Language runtimes, container images, curl builds, and appliances that pin an old CA bundle keep trusting a chain until that bundle is rebuilt. DSCI flagged servers, payment APIs, and devices with frozen trust stores as the residual risk, because the keys had been recoverable since April 2025 and it is not known whether anyone used them against those clients.
eMudhra says seven of the eight intermediates never went into production, that EM DV TLS CA – G3C was live but issued no end-entity certificates, and that the two code-signing issuers never signed software. No certificates from those eight appear in Certificate Transparency logs. The distrust still reached past them, because Chrome dropped the G3 root above the whole hierarchy, not only the broken issuers.
Sixteen Certificates and a Sectigo Cross-Sign
Each of the eight P-384 keys exists in two certificates: one under eMudhra’s G3 chain and one cross-signed in May 2025 under Sectigo’s legacy AAA Certificate Services root. That is 16 certificates on eight secrets. Sectigo did not generate the keys. It signed copies. Because the weakness is in the key itself, each copy had to be revoked on its own. Chrome’s block matches the public key, so it covered both paths. Apple and Microsoft still trusted the AAA root, so the cross-signs were a second path on those platforms until 19 and 20 October.
The ceremony report also covers a second set of 16, and it is a different 16. All 16 B and C series key pairs, eight elliptic-curve and eight RSA-4096, went through the same wrap and unwrap script. The RSA objects still verify against their certified moduli. During re-import, though, private attributes such as the private exponent and the first prime sat in the volatile memory of a Python process on the ceremony host. eMudhra said it could not prove those RSA keys met its own key-protection rules and revoked the certificates tied to them as well.
Jackson wrote that he could not recover the P-384 keys in the A series or the A-1 replacements, even with more effort. Those sat on Entrust modules and were not part of the export test.
The TLS Exit Leaves a US Client Ecosystem Mid-Move
On 22 October 2026 the board voted to leave the global SSL/TLS market, a line of business the 21 October press statement put at around 1 percent of revenue. The company had already begun an exit from global TLS after the keys became public. India Digital Signature Certificate and eSign services, plus emSigner, emCA, PrivaTrust, and SecurePass, run on separate systems and were not part of the global TLS chain.
THE TLS BUSINESS IN FIGURES
- Share of sales: eMudhra put the affected global TLS line at around 1 percent of revenue, while warning of reputational harm it said it could not quantify.
- Certificates to move: The incident page cites roughly 30,800 G3 certificates in a United States client ecosystem that had already been given a path onto the G1 RSA root.
- Outside review: KPMG was appointed on 30 October 2026 as the independent reviewer promised in the 21 October statement.
- G1 clock: Chrome and Mozilla will not trust G1 TLS certificates whose earliest Signed Certificate Timestamp is after 16 November 2026; Apple removes the G1 root on 30 November 2026.
The 30,800 figure is not a count of certificates issued by the eight weak intermediates. Those eight, the company says, issued none. It is the population chaining to the G3 root that browsers pulled after the keys became public. On 5 November 2026 eMudhra said it had already moved emSign TLS customers in every region except the United States, where that large community was still switching, and that it expected the rest to be off the chain before Apple’s date.
Replacement certificates were offered at no charge, including Sectigo-issued alternatives. G1 issuance, the firm said, exists only to finish that move, then stops.
FIPS Came Off and the RSA Keys Sat in RAM
The report’s most uncomfortable pages are not the truncated byte. They are the steps that made a host-side script possible at all. Security researcher Ian Carroll wrote that eMudhra disabled FIPS on the HSM, exported the private key into Python on a computer, and then truncated it on the way back in. “I have to admit I did not realize CAs were running HSMs with the ability to do this,” he wrote. “How many CAs are exporting private keys to computers??”
Mozilla’s follow-up questions, posted on the same ticket, ask what approval process authorised turning FIPS-approved mode off, which other key objects sat on those modules while it was off, and whether that step was shown to the qualified auditor. They also ask why the plaintext export, known to the company from 30 April 2025, was not disclosed until November 2026, and why G2C issuers went into production in May 2026. Those are questions on the ticket, not findings. The ticket remains assigned to eMudhra’s Naveen Kumar ML.
Helme, reading the same filing, said the CA/Browser Forum Baseline Requirements allow export and import under 6.2.6 and set storage rules on the module under 6.2.7, and that the ceremony broke both. The louder failure is simpler than a clause number. A module that can spill a CA private key into a general-purpose Python process is already doing the opposite of the job it is bought for.
WHAT WE KNOW
- The cut: A Python read-back kept the leading byte of eight P-384 private keys generated on 30 April 2025.
- The RSA set: Eight RSA-4096 keys from the same wrap test kept valid signatures but had private material in host memory, so eMudhra revoked those certificates too.
- The production claim: eMudhra says no publicly trusted end-entity certificate was issued from the eight weak intermediates.
WHAT IS UNCONFIRMED
- Prior abuse: The company says it is not aware of anyone recovering the keys before 17 October 2026; non-browser clients that skip revocation cannot prove the same.
- Auditor file: Mozilla has asked whether FIPS-off and host export were documented for WebTrust. That answer is not in the public report.
- G2C timing: Mozilla has asked why G2C issuers were put into production in May 2026. eMudhra has not closed that point on the ticket.
The Chrome Root Program, in a CCADB notice, said it was still not satisfied that the risk stopped at the known weak keys, and that TLS certificates chaining to emSign Root CA – G1 whose earliest timestamps are certificates dated after 16 November 2026, 11:59:59 PM UTC, will not be trusted in Chrome 153 and higher. Older G1 certificates keep working in Chrome and Firefox. On Apple platforms, the 30 November 2026 G1 removal ends trust in every certificate under that root, and eMudhra has said the remaining United States customers should be gone before that date.
-
AUTO2 months agoVolkswagen Sells the Osnabruck Plant to Clear a Board Fight
-
NEWS3 months agoApple’s Foldable iPhone Hinges on Samsung Display Supply
-
BUSINESS3 months agoSaksham Gaur Finds He Is His Own Company’s Client
-
AUTO3 months agoTata Motors’ Iveco Tender Opens as Holders Still Decide
-
TECHNOLOGY3 years agoWhat Is SD on Amazon Prime?
-
BUSINESS3 months agoHyderabad 22-Carat Gold Jumps After a False Bargain
-
GAMING3 months agoOnimusha Way of the Sword Preload Hits a Windows 11 Wall
-
ENTERTAINMENT2 years agoAndy Scott’s Steel Loon Repeats the Kelpies Regeneration Bet
