live wire
▸Snowflake agent evals can pin judge versions; unpinned metrics may roll forward and shift scoresSnowflake▸ThoughtSpot makes Spotter 3 default for new instances; it requires download rights and always shares query data with an LLMThoughtSpot documentation▸MotherDuck adds image classification and object location to prompt_jev(), returning typed SQL outputs with confidence scoresMotherDuck▸BigQuery conversational analytics adds Montréal and Toronto; mixed-region agents still default to globalGoogle Cloud Documentation▸Fabric enables Azure OpenAI Copilot access by default, but OpenAI-subprocessor and cross-region processing/storage switches remain offMicrosoft Learn▸Snowflake Cortex Agents now continue by default when named tools are inaccessible, emitting warnings instead of failing the runSnowflake documentation▸Databricks Genie One can use group memberships to tailor answers, while Unity Catalog remains the access-control boundaryDatabricks documentation▸Fabric data agents can query Power BI models with Read alone; Build and workspace access are unnecessaryMicrosoft Learn▸Fabric data-agent Git configs expose column flags, but selecting a table still includes every columnMicrosoft Learn▸Google Data Agent Kit offers remote Spark kernels, but notebook exports can still move results into local Pandas or ArrowGoogle Cloud Documentation▸Cube says production analytics agents need explicit partial-result states, caller-filtered model search and durable artifactsCube▸Thinking Machines names Bridgewater behind ReViSQL-K2.6, citing 92.97% SC-16 accuracy at 15% of frontier-model costThinking Machines Lab▸AWS says a mismatched data/tool/prompt release cut its Field Sales agent pass rate from 78% to 38% in pre-productionAWS Artificial Intelligence Blog▸NL2SQL Multi-Agent ships evaluation artifacts across BIRD, CSpider, Archer and KDD DataSpace—but no software licenseGitHub▸Snowflake agent evals can pin judge versions; unpinned metrics may roll forward and shift scoresSnowflake▸ThoughtSpot makes Spotter 3 default for new instances; it requires download rights and always shares query data with an LLMThoughtSpot documentation▸MotherDuck adds image classification and object location to prompt_jev(), returning typed SQL outputs with confidence scoresMotherDuck▸BigQuery conversational analytics adds Montréal and Toronto; mixed-region agents still default to globalGoogle Cloud Documentation▸Fabric enables Azure OpenAI Copilot access by default, but OpenAI-subprocessor and cross-region processing/storage switches remain offMicrosoft Learn▸Snowflake Cortex Agents now continue by default when named tools are inaccessible, emitting warnings instead of failing the runSnowflake documentation▸Databricks Genie One can use group memberships to tailor answers, while Unity Catalog remains the access-control boundaryDatabricks documentation▸Fabric data agents can query Power BI models with Read alone; Build and workspace access are unnecessaryMicrosoft Learn▸Fabric data-agent Git configs expose column flags, but selecting a table still includes every columnMicrosoft Learn▸Google Data Agent Kit offers remote Spark kernels, but notebook exports can still move results into local Pandas or ArrowGoogle Cloud Documentation▸Cube says production analytics agents need explicit partial-result states, caller-filtered model search and durable artifactsCube▸Thinking Machines names Bridgewater behind ReViSQL-K2.6, citing 92.97% SC-16 accuracy at 15% of frontier-model costThinking Machines Lab▸AWS says a mismatched data/tool/prompt release cut its Field Sales agent pass rate from 78% to 38% in pre-productionAWS Artificial Intelligence Blog▸NL2SQL Multi-Agent ships evaluation artifacts across BIRD, CSpider, Archer and KDD DataSpace—but no software licenseGitHub
nl2sql.ai
releaseSECURITY RELEASE

LibreDB Studio 0.16.2 fixes a session-boundary failure in its database cache

The flaw reached both ordinary queries and the agent path, showing why read-only execution and connection isolation must be tested separately.

Old cache key vs. patched cache key preventing connection reuse.
AI-generated illustration
By The Tools Desk· Sep 21, 2026the quick take — two AI hosts go live when you do

LibreDB Studio has released 0.16.2 to fix a provider-cache flaw that could let any signed-in user reuse another user’s live database connection. The project’s security advisory says an attacker could supply the victim connection’s ID and receive the already-authenticated provider stored under that key, rather than creating a connection from the credentials in the new request.

That distinction matters for database-agent operators. LibreDB documents a read-only agent that drafts SQL, reads results and writes reports whose claims cite those results. Its README says agent statements run through an audited pipeline and engine-specific read-only execution profiles. But the advisory says the vulnerable cache lookup also existed on the agent path. In other words, read-only SQL controls did not establish whose connection the agent was using.

The cache confused identity with a label

Before the patch, LibreDB cached a live provider using a caller-supplied connection.id. A later request naming the same ID could receive that provider without proving that its endpoint and credentials matched the original. The advisory says the ordinary /api/db/query path required a session but not a privileged role; inline connection data could also claim a seed: ID and bypass the role filter normally applied to administrator-managed seed connections.

The impact was broader than unwanted reads. According to the advisory, the inherited provider carried the original user’s database rights, so a less-privileged Studio account could execute SQL with those credentials. Supplying a different timeout against the same cache entry could also disconnect and evict the victim’s provider.

What 0.16.2 changes

The 0.16.2 release incorporates pull request 1027. The new cache key combines the stored ID with a fingerprint of the database endpoint and a SHA-256 digest of the credentials, using length-prefixed components. The ID remains metadata on the entry for cleanup and lookup operations, while inline requests can no longer claim the reserved seed: namespace. LibreDB says it reproduced both vulnerable paths against the real SQLite driver and converted those reproductions into regression tests, including agent execution profiles.

This is a useful separation of controls for any multi-user database agent. A read-only transaction limits what a correctly selected connection may do. It does not prove that a cache, pool or tool router selected the correct user’s connection in the first place. That boundary needs its own tests: two accounts, the same public connection label, different endpoints or secrets, and assertions that no provider object, result or cleanup action crosses between them.

LibreDB’s advisory recommends upgrading to 0.16.2. It says the only mitigation before upgrading is to trust every account able to reach the instance—an unusually clear signal that this is an isolation fix, not merely cache hardening.

Filed by The Tools Desk. Corrections: desk@nl2sql.ai · Our standards →

comments · 0

    Comments are moderated before they appear. Your email is used once to confirm it is you — never shown, never sold. Corrections and questions get an answer from the desk when we have one.