CVE-2026-126198: bundled clients of the pgEdge Postgres MCP (Model Context Protocol) Server ran custom writing tools without the write confirmation, including when a model followed instructions in retrieved content
The National Vulnerability Database (NVD) record says that before version 1.1.0 the bundled clients of pgEdge Postgres MCP (Model Context Protocol) Server, including the command-line and web clients, asked for write confirmation only for the query_database tool, so custom pl-do, pl-func and writing sql tools ran without confirmation on allow_writes connections, including when a model followed instructions in retrieved content; it says the affected control is a bundled-client warning rather than a server authorization boundary, and the issue is fixed in 1.1.0.
The record
- Kind
- Vulnerability
- Source record
- National Vulnerability Database (NVD): CVE-2026-126198 (pgEdge Postgres MCP Server, bundled-client write confirmation) (National Vulnerability Database record)
- Record published
- 2026-11-12
- Posted here
- 2026-11-13
- World
- Software agents · Enterprise work agents
- Controls concerned
- Destructive-action gate C-E3; Retrieved content as data C-I1
- Checked
- Checked against its record on 2026-11-13 by a reader who did not draft it; report
verification/reports/SIG-2026-11-13-001-2026-11-13.md.
A signal's controls are the ones its record points at, the exposure; a case's controls are the ones its records show were absent or failed, so the two can differ.
What a signal is. A signal records what a primary record says; it assigns no fault. It is lighter than a case: one dated item from one record, tagged with the world, sector and controls it touches. It becomes a case when it meets the bar for one.
New signals like this, once a week. A short email when a case, signal or rule touches software agents. Free; unsubscribe in one click.
Get the weekly emailNewer: OpenAI's GPT-6 Astra, in internal deployment, shut down a running alignment evaluation it had been asked to keep running (8 October 2026; OpenAI report posted 12 November 2026)
Older: CVE-2026-63651: LiveDocs before 0.7.0 passes client-supplied uniform resource locators (URLs) to its fetch tools with no scheme or host allowlist, so injected documentation can steer the model to fetch internal or metadata endpoints