Fetching from the wire…
Public story · 2026-09-11 · high
Simon Willison paired Fable 5.1, GPT-5.6 and GPT-6 Astra with two human reviewers to find a case-insensitive table name bug that had slipped past solo review.
Why now: Willison published the releases and the security writeup on September 10.
Simon Willison released Datasette 1.0a39 and 0.65.4 on September 10, fixing three permission bugs found in a security audit. The audit paired Fable 5.1, GPT-5.6 and GPT-6 Astra with human reviewers Sevban Dönmez and Alex Garcia.
Datasette runs as a public tool over SQLite databases that often mix public and private tables in one instance. A permission bug there can expose private data to anyone who can query the public side.
Permission checks ignored SQLite's case-insensitive table names, letting a differently-cased table name slip past an access rule. FTS index tables were readable with no permission check at all. Restricted actors could create API tokens.
Willison calls the bugs "very subtle." A case-insensitive table name reads correct at a glance. It only breaks under SQLite's own matching rules.
The writeup doesn't say how the audit was structured. It's unclear whether the three models reviewed the same code independently or split it among them.
Upgrade now if you run a mix of public and private tables on one instance, rather than waiting for the next routine release.
Each link below shares sources, entities, or timing with this story.
OpenAI published "Research acceleration: the view inside OpenAI" on September 6 with numbers no lab has put in public before (OpenAI). As of mid-August, the research organization uses 3.1 agent-workdays of effort for every workday of human labor. It says it reached its interna...
The Astra coverage went to price and context window. The number that changes how I'd deploy it went into a system card nobody read. Artificial Analysis measured GPT-6 Astra's hallucination rate on AA-Omniscience at 51% at max effort, against 92% for its predecessor. Accuracy w...
Simon Willison shipped the fix August 6 and back-ported to 0.65.3: instances serving both public and private tables from one database let public-table users inject SQL and gain read access to private tables, defeating the permission system even with execute-sql disabled. Willi...
Prompted by Julia Evans admitting on July 17 that she still can't read query plans, Willison had Fable build a tool that runs arbitrary SQL against a SQLite database and renders both EXPLAIN QUERY PLAN and the lower-level EXPLAIN bytecode with per-line plain-English annotation...
Anthropic shipped Fable 5 on June 9. Willison spent ~5.5 hours stress-testing it: slow and expensive, but it handled everything he threw at it, including agentic coding. (Simon Willison) The tell that it's a real working model and not a benchmark queen: because it post-dated A...
The writeup covers a 4.0 stable adding database migrations, nested transactions via db.atomic(), and compound foreign keys, with rc2 work costing about $149.25 in Fable usage. The transferable technique isn't "have the model write code." It's running a model as a dedicated rev...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.