SShortSingh.
Back to feed

Why AI Data Agents Need More Than RBAC to Prevent Sensitive Data Leaks

0
·4 views

Role-based access control (RBAC) remains essential for enterprise data security, but AI analytics agents introduce a new vulnerability known as the inference gap, where restricted information can be derived from permitted data. For example, a user denied access to employee salary figures could still calculate average compensation by combining authorized fields like department total cost and headcount. Traditional authorization models govern access to database objects but do not account for what an AI agent can logically infer from those objects. To address this, experts recommend semantic authorization, where policies are defined at the business-concept level and evaluated before SQL is generated, not just at query execution. This approach ensures that authorization constrains the reasoning context of an AI agent early in the pipeline, closing the gap between what a user can access and what they are permitted to learn.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Vocabulary Caps Hurt Small Language Models, New Experiment Finds

A developer running experiments on small transformer models tested whether capping vocabulary size — a feature associated with the TinyStories paper — improves model performance on a cs.CL abstracts corpus. Three vocabulary caps (1,500, 4,000, and 8,000 types) were tested on an identical 4-layer model with 846K training tokens, scored only on token positions common to all arms to ensure fair comparison. The 1,500-token cap scored 4 accuracy points lower than larger vocabularies and produced severely degraded text, with over a third of output tokens replaced by an unknown-word placeholder. Unlike TinyStories, where simple words carried full meaning in children's stories, capping vocabulary on a technical corpus strips content words and forces the model into repetitive, skeleton-like outputs. The experiment concludes that TinyStories' success came from its domain and writing style, not from the vocabulary restriction itself.

0
ProgrammingDEV Community ·

Investing Knowledge Graph Series: Why Entity Resolution Breaks When Merging Multi-Source Data

A developer building an investing knowledge graph shares insights from two real-world conversations that shaped the final part of their series. One discussion involved a financial news aggregation system merging data from English and Chinese sources, such as matching 'Tesla, Inc.' with '特斯拉', highlighting challenges in cross-language entity blocking. The other involved a KYC sanctions screening system, where the stakes of a false negative — an entity wrongly cleared — are far more serious than in a typical knowledge graph. While the core entity resolution architecture transfers across both use cases, the author cautions that model thresholds and training data must be domain-specific, especially for compliance tools. Cross-language name matching and alias generation during ingestion are identified as key technical hurdles that standard string normalization alone cannot solve.

0
ProgrammingDEV Community ·

Google Apps Script Gets Python and Bash Execution via Cloud Run Sandboxes

A developer has demonstrated how Google Cloud Run Sandboxes, powered by gVisor isolation technology, can be connected directly to Google Apps Script to overcome the platform's built-in limitations. Standard Apps Script environments restrict users to JavaScript on the V8 runtime with a strict six-minute execution timeout, blocking native Linux binaries and Python libraries. By routing tasks through Cloud Run Sandboxes, developers can now run Python data science tools like Pandas and Bash commands in 200–450 milliseconds with zero idle cost. The architecture uses zero-trust micro-isolation, meaning untrusted or AI-generated code executes in ephemeral containers without exposing the host environment. The approach is positioned as a lightweight alternative to LLM-based agents for deterministic Workspace automations that do not require conversational inference or token consumption.

0
ProgrammingDEV Community ·

Free Browser Tool Generates 47-Tile Autotile Sets from Just Two Source Tiles

A developer has released a free, browser-based tool called 47-Tile Terrain Autotile Maker that automatically generates a complete autotile set from two user-selected ground tiles, eliminating the need to draw every edge, corner, and transition manually. The tool works by reducing 256 possible neighbor combinations to the 47 valid ones that can realistically appear on a map, matching the tile format expected by both Godot's terrain system and Unity's Rule Tile. Users can choose from six built-in terrain pairs, select tile sizes between 16 and 64 pixels, and customize boundary styles, corner roundness, and edge roughness before downloading. Exports include a PNG tile sheet at up to 8x resolution, a sample map image, and a JSON manifest with neighbor masks and engine-specific conditions for Godot and Unity. The tool runs entirely in the browser with no sign-up required and no file uploads, and includes 57 automated checks to ensure output colors are drawn only from the two source tiles.