Closing the Binary Blind Spot: Domain-Knowledge Injection for LLM Malware Triage in npm Packages
Summary
Injecting domain knowledge into large language models resolves the binary blind spot to enhance automated malware triage across npm packages. This approach strengthens software supply chain defenses against the rising volume and sophistication of attacks targeting package registries.
Summary generated by brickster.ai. For the full article, follow the source link above.
Topics
More from Databricks Blog
Transactions are Generally Available, advancing warehousing on the Lakehouse
Multi-statement transactions are now generally available for Delta Lake and Apache Iceberg tables on Databricks, bringing native atomic guarantees to multi-table workflows. Backed by row-level concurrency and Unity Catalog governance, this capability allows practitioners to run critical warehouse workloads directly on open lakehouse data without maintaining separate warehouse systems.
How our AI Research team uses AI Runtime
Databricks AI Research uses AI Runtime to train and fine-tune models across hundreds of GPUs, supporting an 8x surge in training jobs fueled by agent-driven research. Built-in GPU reliability, fast job launches, and native integrations with MLflow and Unity Catalog accelerate these automated experimentation workflows at scale.
Stripe Data Pipeline now syncs real-time Stripe data to Lakebase Postgres
Stripe Data Pipeline now streams real-time Stripe data directly into Databricks Lakebase Postgres with no code required, eliminating the need to build and maintain custom API or webhook integrations. Applications can now query always-current Stripe data using simple SQL to gate features, display live billing details, and trigger personalized offers.
Reducing the noise: How Atlassian built an OCSF security lakehouse on Databricks
Atlassian normalized 44 billion daily events into an OCSF security lakehouse on Databricks, cutting ingestion costs by 80% and accelerating queries by up to 95%. The unified architecture now powers over 500 production threat-detection signals while driving a 60% increase in security analyst productivity.
Geospatial queries are up to 38X faster with Lakehouse//RT
Lakehouse//RT now supports spatial SQL and spatial joins directly on governed lakehouse data, delivering up to 38X faster query performance than traditional data warehouses under 50 concurrent users. This eliminates the need for a separate geospatial serving layer while bringing real-time latency, Unity Catalog governance, and more than 100 spatial functions to your architecture.
Sizing Lakebase Search: lessons from a real app and 4 million passages
Sizing the Local File Cache to match the vector index keeps up to 98% of pages in memory and drops warm vector latency to single-digit milliseconds directly inside Lakebase Postgres. Reducing the search candidate pool to 200 cuts page reads by two-thirds while preserving 90.5% of top-10 results, with an always-on 0.5 CU floor providing an optimal production baseline for just $31 to $40 per month.