Low severity

Datasette 1.0a43 fixes flaw exposing private stored query titles and SQL

The maintainer says the bug leaked what a private query is called and what it contains, but did not let anyone run it.

By Mr Elite Published 5 min read

Severity
Low
Status
Confirmed by the organisation
Sources
1
Last checked

The short version

  • Simon Willison released Datasette 1.0a43 on 15 November 2026.
  • He said private stored queries could be exposed, revealing their title and SQL.
  • He said the flaw did not allow the query to be executed.
  • He published a GitHub advisory for it and said the release has another security fix.
  • No CVE number, affected version range or exploitation reports were given in his post.

What happened

Simon Willison, creator of the Datasette data-publishing tool, released version 1.0a43 on 15 November 2026 with a security fix for private stored queries that could be exposed. Datasette is an open source project for publishing and exploring databases through a web interface.

In the release notes on his blog, Willison said the release contains a couple of security fixes. He singled out one as deserving a GitHub advisory, the formal notice projects use to describe a vulnerability and its fix.

He described the problem this way: private stored queries could be exposed, giving access to the title and SQL of the query. He said it did not allow the query itself to be executed. A stored query is a saved SQL statement that Datasette presents under a name, so people can run it without writing SQL.

Willison did not say in the post what the second security fix covers. He also did not give a CVE number, a severity score, the earliest affected version or any sign that the flaw was used against real deployments.

Most of the release notes cover other work, including streaming support for output renderers and fixes contributed by several outside developers. Willison said the 1.0 milestone is now 53% complete.

Why it happened

Willison's post states the outcome but not the code-level cause. He said private stored queries could be exposed, and that the exposure covered the title and SQL text only.

Which page, endpoint or listing leaked the details, and who could see them, is not stated in the post. The GitHub advisory he mentions may hold more detail, but its contents were not part of the material available for this story.

How this kind of failure works

Flaws like this are usually authorisation gaps. A system checks permission in the place where an action happens, such as running a query, but forgets to check it in the places that merely describe the action, such as lists, search results or error messages.

Think of an office filing room where each drawer is locked. If the label on every drawer is printed in a public corridor directory, a visitor cannot open anything, but they now know what is kept there and how it is organised.

For a stored query, the SQL text is more than a label. It can name tables and columns, show how data is joined and filtered, and sometimes contain fixed values that someone typed in for convenience. That gives an attacker a map of the database for later attempts.

Developers can miss these paths because each feature is built and tested on its own. Any new page or output format that lists queries has to repeat the same visibility rule, and one that does not becomes a side door.

Who is affected

Willison's post does not list affected versions. Anyone running Datasette with private stored queries, meaning queries restricted to certain users, should treat the issue as relevant until they have read the advisory.

To check your own exposure, look at which stored queries you have marked private and whether any appear for a signed-out visitor. Compare your installed version with the advisory's list of affected releases once you have it.

In practice, the exposure is limited to knowledge. Willison said the query could not be run. The risk grows if the SQL text names sensitive tables or contains embedded values, so the severity here depends on what your queries say.

How to defend against it

Tick each step as you check it. Your progress is saved on this device.

What to learn from this

The concept: Authorisation on metadata, not just actions

Protecting an action is only half the job. The names, descriptions and source text of protected things can also be sensitive, and every place that displays them needs the same permission check as the thing itself.

The pattern appears widely. Search tools can show titles of documents a user cannot open, cloud storage listings can reveal file names, and error messages can reveal table names. Each is a leak of information about the protected item.

Apply the same access rule to the description of a protected thing as to the thing itself.

Quick check

Sign in to earn 10 XP for a right answer

A private stored query's SQL can be seen but not run. Why can this still matter?

Timeline

  1. 15 November 2026, 23:20 UTC Simon Willison publishes the Datasette 1.0a43 release notes on his blog.
  2. With the 1.0a43 release A GitHub advisory is issued for the private stored query exposure, according to Willison.
  3. With the 1.0a43 release Willison says the release includes a second security fix, which he does not describe.

What is not known yet

  • The CVE number, if one was assigned.
  • Which earlier versions are affected.
  • What the second security fix covers.
  • Whether anyone exploited the flaw before the fix.
  • Exactly which pages or endpoints leaked the query details.

Questions people ask

What did the Datasette 1.0a43 security fix change?

Willison said private stored queries could be exposed, revealing their title and SQL. The release fixes that. He also said the release has a second security fix but did not describe it.

Could attackers run private queries because of this bug?

Willison said the flaw did not allow the query to be executed. It exposed only the title and SQL text. Whether other weaknesses exist is not addressed in his post.

Do I need to upgrade Datasette?

If you use private stored queries, read the GitHub advisory and upgrade once you have confirmed the affected versions. 1.0a43 is an alpha release, so test it before using it in production.

Has the flaw been exploited?

Willison's post does not say. It gives no sign of exploitation, and no CVE or severity rating. Check the advisory for later updates.

Sources

Know someone who should see this?

Keep reading