NVD published CVE-2026-116779 on November 5: FreeSWITCH before 1.11.3, CVSS 9.8 (AV:N/AC:L/PR:N/UI:N), and the record says exploitation has been observed in the wild. SecurityWeek is reporting it alongside an IBM Aspera Faspex bug. The Aspera one can wait. This one is a telephony switch executing commands off a text message.
Here’s the mechanism, per the NVD entry. When an inbound SIP MESSAGE arrives, sofia_presence_handle_sip_i_message() splits the To user on a plus sign and treats the left side as a chat protocol selector. Pick api and it reaches switch_api_execute(). That means a peer whose MESSAGE is accepted can run FreeSWITCH API commands, including ones that hit operating-system command sinks, with the daemon’s privileges. The record says inbound chat is on in the stock internal and external profiles, and any authenticated profile account is enough. Stock directory users without password material can remove even that barrier in affected configurations. Configuration and credential disclosure, call control, and command execution all follow.
Read that again. The To header is attacker-controlled input, and the part before the + picked which handler ran. Somebody built a feature where an address field is a function name. I have watched people do this with finger and with CGI query strings, and it ends the same way every time. The PR text confirms the shape: the api proto “runs the address as a FreeSWITCH API command.”
The fix landed as PR #3135, commit 7015f6a. It adds a per-profile enable-chat-api-proto parameter. Off unless set. A MESSAGE selecting api on a profile without it gets a 403, and the log records the source and the requested command. It also makes the proto compare case-insensitive, matching the chat interface registry, which is built with switch_core_hash_init_nocase(). Fair enough: a case-sensitive gate in front of a case-insensitive lookup is a bypass waiting for a bored teenager. The param ships commented out in the four vanilla profiles and the sofia.conf.xml sample, with a note that enabling it runs FreeSWITCH API commands and should be for trusted peers only. NVD lists 1.11.3 as the fixed version.
I’ll give them this: the patch returns early rather than jumping to the end label, which would have answered 200/202 on top of the 403. The code comment says so. Somebody read their own error path, which is more than most CVE fixes can claim.
What to do:
- Find every FreeSWITCH box, including the one the VoIP contractor left behind.
fs_cli -x versiontells you what you’re running. - Below 1.11.3, upgrade. If you can’t today, turn off inbound chat on the internal and external profiles (
enable-chatis the existing profile flag the patch touches). I’d verify that against your own config before trusting my memory of it. - Assume the SIP ports are scanned already. Check logs for MESSAGE requests with an
api+prefix in the To user, and rotate anything the daemon could read: gateway credentials, directory passwords, ESL secrets. - SIP on 5060 and 5080 does not need to face the whole internet. ACL it to your carriers.
Same family of mistake as the Flowise reset token from yesterday: the dangerous path ships on by default and the safety gets bolted on after the first exploit. More of this in the kernel and security hub, and if you want to know what is actually bound to your sockets, start with what is actually listening. Go patch.
