AWS pgcollection CVE-2026-96883

Requires the pgcollection extension (versions 2.0.0 through 2.1.1) to be installed in a target PostgreSQL database, and an authenticated remote database user (PR:L) with the ability to execute SQL that invokes pgcollection's collection value-retrieval or icollection_to_array array-conversion functions. Exploitation is limited by that authentication requirement (a valid, even low-privileged, database role is needed) and by the extension being a non-default, explicitly-installed add-on - databases without pgcollection or restricting SQL execution to trusted roles are not exposed. No user interaction beyond the attacker's own SQL is required (UI:N).

CVSS 4.0 base score 8.7 (HIGH) with vector AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H, indicating a network-reachable, low-complexity, low-privilege attack yielding full confidentiality, integrity and availability impact on the vulnerable system - consistent with the CWE-843 type-confusion RCE described. The critical qualifier is PR:L: exploitation requires an authenticated database role able to execute SQL, so this is not a pre-auth internet worm but rather a privilege-escalation-to-OS-code-execution primitive (an authenticated PostgreSQL user gaining code execution as the postgres OS user, which typically breaks out of the SQL sandbox and compromises the whole database host). No EPSS score, no CISA KEV listing, and no public POC are present in the provided data, so the exploitation status is 'no public exploit identified at time of analysis.' The vendor (AWS/AMZN) has published an advisory (2026-118-aws), a GitHub Security Advisory (GHSA-g539-cj32-hv6r), and a fixed release (v2.1.2), and the release notes confirm the exact root cause (over-permissive can_coerce_type() replaced with exact/binary-coercible checks, plus a polymorphic icollection_to_array return-type fix). Real priority is moderate-to-high for the specific population running this extension: the impact is severe (OS-level RCE) and the fix is well-characterized, but the authenticated prerequisite and the niche footprint of the extension bound widespread mass-exploitation risk. No signals conflict; the main gap is the absence of exploitation-probability telemetry (EPSS) and POC data.

An attacker who already holds a low-privileged but valid PostgreSQL login on a database with the pgcollection 2.0.0-2.1.1 extension installed submits crafted SQL that invokes collection value-retrieval or icollection_to_array conversion functions with mismatched type metadata. The type confusion causes datumCopy to operate on a datum representation inconsistent with the assumed type, corrupting memory and ultimately executing arbitrary code as the postgres OS user, breaking out of the SQL sandbox to compromise the host. No public exploit code is identified at time of analysis; the attack is remote and low-complexity but requires authenticated database access.