Adafruit published CircuitPython 10.3.2 on 11 November 2026 at 19:19 UTC. It is a stable bugfix release with one change. The release notes list a single highlight since 10.3.1: "Remove compromised certificates from root certificates bundle." The underlying change, credited to @dhalbert in PR #11838, updates the root certificate bundle to remove compromised certificates from eMudhra emSign and AAA Certificate Services.
A one-line changelog entry would be easy to skip, but this one matters. Until this release, a CircuitPython board making an HTTPS connection would have accepted a server certificate signed with any of eight intermediate CA keys that were publicly shown last month to be guessable. The reason is not a CircuitPython coding error. It comes from the way the TLS library on these boards handles a revoked certificate. The rest of this article covers what broke and why.
What happened to eMudhra's intermediates?
According to ID Tech's 22 October report, citing Mozilla's incident report, a researcher reported on 17 October that eight intermediate CAs chaining to eMudhra's emSign ECC Root CA - G3 had compromised keys. Mozilla independently recovered the keys in under two hours. ID Tech reports that each key was a number smaller than 256, when a key of this type is meant to be a random number more than 100 digits long. Six of the eight intermediates were authorities for website certificates and two for code-signing certificates.
eMudhra revoked all eight on 18 October and said a flawed script used at a key generation ceremony in April 2025 was to blame. Mozilla and Google are removing trust in emSign ECC Root CA - G3. ID Tech also reports that India's Ministry of Electronics and Information Technology (MeitY) has set up a team to investigate. It notes that Mozilla is removing trust in the entire root, which protects devices and software that cannot check whether individual certificates have been revoked.
Where does Sectigo come in?
Sectigo had cross-signed the compromised eMudhra intermediates. In its advisory, Sectigo says that on 19 October 2026 it revoked the cross-certificates for the eight compromised eMudhra CAs, and that it also revoked additional intermediate certificates it had cross-signed to eMudhra, ending the cross-signing relationship. Sectigo says none of its own systems or keys were compromised, and that Chrome and Mozilla removed the compromised CAs from their root stores.
For a desktop browser that would mostly close the issue, since the cross-certificates are revoked and the root is being distrusted. A client that does not check revocation is in a different position.
Why did CircuitPython still trust the leaked keys?
The adafruit/certificates PR #14, merged by dhalbert on 10 November, explains the problem. Its description says tools/extra.pem, a supplementary file used alongside the Mozilla root list, had added AAA Certificate Services (Comodo) back after Mozilla removed it in 2025. The PR fixes issue #13, which reported this.
Sectigo cross-signed the eight compromised eMudhra intermediates under AAA Certificate Services (tracked as Mozilla bug 2083449). The PR notes that those cross-signs are revoked, but mbedtls does not check revocation, so both bundles accepted certificates signed with the leaked keys. Any client that trusts AAA would accept them.
In practice, anyone holding one of the recovered keys could have issued a certificate that a client trusting AAA and not checking revocation would validate, because the chain ran from the compromised intermediate through the Sectigo cross-sign to a root the device still trusted. ID Tech notes that anyone who holds an authority's private key can create certificates for any website. The revocations did not help clients that never check them.
What exactly changed in the bundle?
According to PR #14:
- Removed: AAA Certificate Services and, as separate cleanup unrelated to issue #13, the Starfield Class 2 Certification Authority root.
- Added: Sectigo Public Server Authentication Root E46 and Sectigo Public Server Authentication Root R46.
- Compatibility note: AAA had been added back for github.com, which now chains to USERTrust ECC. Some servers, such as softpedia.com, send R46 cross-signed by AAA; the PR says those still verify with R46 in the bundle.
- Upstream parallel: the PR says ESP-IDF made the same changes to AAA and Starfield Class 2 in espressif/esp-idf#19282.
- Backport: the new bundle reaches CircuitPython through adafruit/circuitpython PR #11838, "[10.3.x] lib/certificates: update to 20261110 to remove AAA Certificate Services". The release notes credit #11838 as the change shipped in 10.3.2.
The new Sectigo E46 and R46 roots matter because removing a widely used legacy root can break legitimate sites that still chain through it. The PR reports that its site-coverage test showed no new failures in either bundle. The sources do not list which other hosts, if any, will stop validating without AAA or Starfield Class 2. Anyone whose devices call less common APIs should test those endpoints after upgrading.
Which boards are affected?
Any CircuitPython board that makes TLS connections with the bundled roots. The 10.3.2 release notes list these as stable ports:
- atmel-samd: Microchip SAMD21, SAMx5x
- cxd56: Sony Spresense
- espressif: ESP32, ESP32-C2, C3, C5, C6, C61, H2, S2, S3
- nordic: nRF52840, nRF52833
- raspberrypi: RP2040, RP2350
- stm: STM32F4 chip family
Alpha ports include espressif ESP32-P4, broadcom (which the notes say will be removed in CircuitPython 11), zephyr and Silicon Labs MG24.
Not every board on that list has a network interface. The exposure is greatest on boards that do HTTPS directly, which in practice means Wi-Fi-equipped Espressif parts and RP2040/RP2350 boards with wireless modules. A board that never opens a TLS socket gains nothing from the update, but 10.3.2 is still the current stable release.
Does this affect ESP-IDF and Arduino-ESP32 projects too?
The adafruit/certificates PR states that Espressif made the same AAA/Starfield change in espressif/esp-idf#19282. The sources do not say which ESP-IDF release includes that change or how it reaches downstream frameworks. Developers who build ESP32 firmware directly on ESP-IDF should check whether their pinned IDF version includes it. They should not assume that a CircuitPython update covers their other firmware.
What It Means for Makers
Who it is for: anyone running CircuitPython on a networked board who talks to HTTPS endpoints, such as IoT dashboards, MQTT over TLS or REST APIs. Anyone with a fleet of networked boards on old firmware has the most devices to update.
What it replaces: 10.3.2 supersedes 10.3.1. The only listed change is the certificate bundle update, so the upgrade carries little regression risk beyond the root changes described above.
What it fixes: it removes a trust path through which a certificate signed by any of the eight leaked eMudhra intermediate keys would validate on the device. Because mbedtls does not check revocation, the fix was to remove the root itself. Revocation by Sectigo and eMudhra did not protect clients that do not check it.
The wider lesson: where the TLS library does not consult revocation, the root bundle is effectively the trust control, so an incident that revocation would handle elsewhere needs a firmware release. Devices that never get firmware updates keep trusting the old roots until they are updated.
Availability: CircuitPython 10.3.2 is a stable release, published on GitHub on 11 November 2026. The release notes do not state a price.
Sources
- CircuitPython 10.3.2 release notes (adafruit/circuitpython)
- adafruit/certificates PR #14: Remove AAA Certificate Services and Starfield Class 2 roots; add Sectigo roots
- Sectigo: Advisory on compromised eMudhra intermediate certificates
- ID Tech: Mozilla and Google Drop eMudhra Root Certificate Over Weak Keys as India Opens Probe