Security updates 1.0a39 and 0.65.4 released for Datasette

Simon Willison announced two new security patch versions of the open‑source tool Datasette on 11 September 2026.

Version 1.0a39 targets the current alpha series while version 0.65.4 updates the stable 0.65.x branch.

Both releases address vulnerabilities that could affect deployments exposed on the public internet.

Willison specifically warns operators who serve a mix of public and private tables to apply the patches promptly.

What the patches fix

The updates resolve subtle bugs uncovered during an extensive security audit.

The audit identified issues that might allow unintended data exposure when public and private datasets coexist in a single Datasette instance.

By correcting these flaws, the patches aim to prevent accidental leakage of private information through publicly reachable endpoints.

AI‑driven audit methodology

Willison, along with contributors Sevban Dönmez and Alex Garcia, conducted the audit using frontier language models.

The models employed were Claude Fable 5.1, GPT‑5.6, and GPT‑6 Astra.

These “coding agents” examined the codebase, generated test cases, and highlighted edge‑case behaviours that human reviewers might miss.

The collaboration with multiple models helped surface “very subtle bugs,” according to Willison.

Willison notes that the experience will shape future development, stating that security audits by frontier models will become a regular part of the workflow.

Collaborative fix workflow

Alex Garcia introduced a split‑work approach that Willison found “extremely productive.”

In practice, one developer wrote automated tests that reproduced the identified issue.

The other developer then implemented the corresponding code fix.

This two‑person review, combined with independent model analyses, ensured that each vulnerability received scrutiny from both humans and AI agents.

The pair worked in a shared private repository, allowing rapid iteration and clear audit trails.

Willison emphasizes that this method guarantees “two separate humans had eyes on each of the issues, in addition to our coding agents running different models.”

After a week of joint testing and review, the fixes were merged into the released patch versions.

Implications for Datasette users

Operators of public‑facing Datasette services are now advised to upgrade to the new versions without delay.

Applying the patches reduces the risk of data leakage in environments where public and private tables are intermingled.

The announcement also signals a broader shift toward integrating large language model audits into open‑source security practices.

Developers who rely on AI‑assisted code review can look to this workflow as a concrete example of how to combine human oversight with model‑generated insights.

Willison’s blog post lists the release as a “security release,” underscoring the critical nature of the updates.

Users can verify the patches by checking the version numbers—1.0a39 for alpha and 0.65.4 for the stable line.

Further details, including the full list of addressed issues, are available in the repository linked from the blog.

Why This Matters: applying the patches secures public‑facing Datasette instances that combine public and private tables against the subtle vulnerabilities identified by the AI‑driven audit.