Wednesday, August 05, 2026

pgAdmin 4 version 9.17 is now available, addressing seven distinct security vulnerabilities identified in the CVE-2026-17346 through CVE-2026-17566 range. This update also delivers 28 additional bug fixes and introduces new features such as logging authenticated user identities in HTTP access logs. Key usability improvements include an opt-in Gateway API HTTPRoute template for Helm charts and a configurable row count cap for the View Data action to handle large tables more efficiently.

• Patch immediately to resolve seven critical security vulnerabilities in pgAdmin 4.
• Enable HTTP access log identity tracking for better audit trails and security monitoring.
• Use the new Gateway API HTTPRoute template in Helm charts as an alternative to Ingress.
• Configure row count limits in View Data to prevent performance issues on large tables.

Percona researchers investigating memory allocator performance in Percona Server for MySQL discovered unexpected memory consumption patterns linked to stored procedures. The findings emerged during benchmarking and reveal details that warrant deeper explanation beyond the original performance comparison goals. These results highlight specific memory behaviors that may impact server resource planning.

• Stored procedures exhibit unique memory allocation characteristics in Percona Server for MySQL.
• Memory allocator choices significantly influence observed performance and consumption metrics.
• Benchmarking stored procedures requires accounting for these specific memory overheads.
• Review Percona's detailed analysis to understand the root causes of these memory effects.

PostgreSQL 18 adds a Grand Unified Configuration (GUC) named extension_control_path, allowing extension control and SQL files to reside outside the server's installation directory. This change enables packaging extensions as independent OCI container images that can be mounted at runtime via Kubernetes ImageVolume or Docker volumes. The approach supports decoupling extension versions from the core server image, though its effectiveness depends on the specific extension's architecture and dependencies.

• New GUC extension_control_path decouples extension files from the PostgreSQL server binary tree.
• Enables packaging extensions as standalone OCI containers mounted at runtime in Kubernetes or Docker.
• Reduces server image rebuilds by allowing independent versioning of extensions like pgvector.
• Decoupling benefits vary; complex extensions with deep system integration may not gain value.

AWS Transform for Windows modernization now allows offline source transformation, enabling customers to migrate Microsoft SQL Server databases to Amazon Aurora PostgreSQL without live connections. The service uploads DDL files to assess complexity, converting storage objects via AWS DMS and stored procedures using an agentic interface. This facilitates direct modernization of legacy .NET applications by decoupling schema analysis from active database access.

• Offline migration removes dependency on live SQL Server availability during schema assessment.
• DDL upload enables early complexity analysis for stored procedures and storage objects.
• Agentic tooling assists in converting SQL Server code to Aurora PostgreSQL equivalents.
• Streamlines legacy .NET app modernization by separating schema work from cutover.

The pgBackRest community has released version 2.59.0, a reliable backup and restore solution designed for large databases. This update includes support for parallel backup and restore operations, block incremental backups, and multiple compression types. It also features malware and ransomware protection, per-file checksums, and encryption options.

• Upgrade to pgBackRest 2.59.0 for improved backup performance and reliability.
• Leverage block incremental backups to reduce storage usage and backup times.
• Implement malware and ransomware protection for enhanced security.
• Utilize multiple compression types and encryption for optimized storage.
• Ensure per-file checksums are enabled for data integrity verification.

AWS has general availability for a built-in server-side Web Search tool in Amazon Bedrock, enabling OpenAI GPT-5.4 through 5.6 models to ground responses in current web data. This feature allows data residency within the AWS environment with zero data egress, eliminating the need for third-party search providers. Engineers can now enable this capability via a single API parameter, removing the overhead of vendor onboarding, separate billing, and additional security reviews.

• Native web grounding for OpenAI GPT-5.4/5.5/5.6 models via a single API parameter
• Zero data egress ensures responses stay within secured AWS environment
• Eliminates third-party vendor onboarding, API key management, and orchestration
• Removes need for additional compliance reviews for external search providers

SIRIN is a new toolkit and interactive web UI designed to detect contextual hallucinations in retrieval-augmented, agentic, and memory-grounded LLM systems. It unifies three detection paradigms—representation probing, uncertainty estimation, and judge-style verification—along with pre-generation query answerability assessment. The tool supports both white-box and black-box settings, enabling response- and span-level inspection through a centralized configuration and evaluation pipeline.

• Unifies representation probing, uncertainty estimation, and judge-style verification in one interface.
• Supports live analysis of context-query-answer triples with hallucination scores and span highlighting.
• Enables span-level inspection for precise identification of unsupported information in LLM outputs.
• Operates in both white-box and black-box settings for flexible integration into existing pipelines.

JudgeArena addresses the fragmentation in LLM-as-a-judge evaluation by unifying major benchmarks like AlpacaEval and Arena-Hard under a single open-source interface. The framework allows for swappable judge models and comprehensive metadata logging to improve transparency and reproducibility. This enables systematic studies on how design choices such as prompts, benchmarks, and inference backends impact model quality conclusions.

• Unifies AlpacaEval, Arena-Hard, MT-Bench, and m-Arena-Hard into one interface
• Supports swappable judge models to isolate variable impacts on evaluation results
• Logs comprehensive metadata to ensure transparency and reproducibility
• Enables systematic analysis of how benchmark and prompt choices affect rankings

CoT-Core addresses the high computational cost of LLM evaluation by introducing a training-free core question selection framework. Unlike prior methods that rely on heavy historical logs or suffer from surface-level lexical bias, this approach prompts LLMs to generate zero-shot Chain-of-Thought reasoning paths. These trajectories are projected into a latent space to cluster questions based on intrinsic logical equivalence rather than superficial wording.

• Eliminates cold-start bottlenecks associated with methods like Item Response Theory.
• Reduces evaluation overhead by clustering tasks on underlying logic, not just vocabulary.
• Operates without additional training, leveraging zero-shot CoT for clustering.
• Mitigates surface lexical bias to better capture the reasoning manifold of tasks.

An arXiv study analyzing 26,804 pairwise judgments from over 736 clinicians reveals that pairwise preference is a weak indicator of clinical safety in large language models. The research, sourced from the MOOVE platform, shows that models favored in head-to-head comparisons can still generate substantial rates of unsafe or misleading content. This disconnect suggests that relying on ranking signals alone may mask critical safety failures in clinical AI applications.

• Pairwise rankings fail to capture safety-critical errors, allowing unsafe models to rank highly.
• MOOVE data uses discrete [-2, +2] rubric scores to flag unsafe content, offering better granularity.
• Clinical AI evaluation must move beyond preference rankings to explicit safety rubric scoring.
• High preference scores do not guarantee absence of misleading or dangerous medical advice.

ARCHead addresses the bottleneck of retaining unquantized LM-heads in weight-only quantized LLMs by replacing dense BF16 projections with a packed structure. It combines a quantized low-rank core, group-wise INT4 residuals, and a low-rank correction derived from activation metrics. On Qwen3-8B-Base, this approach cuts head storage to 25.6% of the original while maintaining a relative perplexity of 1.007, significantly outperforming naive INT4 quantization.

• Packs LM-heads using low-rank cores and INT4 residuals, eliminating dense BF16 storage.
• Achieves 3.7-3.9x storage reduction with negligible perplexity degradation (1.007).
• Activation-derived metric fitting corrects perturbations often seen in naive quantization.
• Outperforms storage-matched naive INT4, which suffers 1.14-1.16x perplexity increase.

AWS has extended its serverless model customization feature in Amazon SageMaker AI to support full fine-tuning for over 25 open-source models. This update allows users to update all model parameters, going beyond parameter-efficient methods like LoRA. The capability enables deeper adaptation for complex tasks requiring specialized reasoning or internalization of proprietary domain knowledge.

• Full fine-tuning is now available serverless, eliminating infrastructure management overhead.
• Supports 25+ open-source models including Llama, Gemma, Qwen, and Nemotron.
• Enables deep adaptation for complex output formats and specialized reasoning patterns.
• Ideal for internalizing large proprietary datasets beyond surface-level style adjustments.

This paper addresses the difficulty of adapting LLM agent behaviors to individual users without expensive fine-tuning. It proposes FABLE, a small, factorized policy model that sits outside the main agent to learn execution decisions from scalar feedback. FABLE separately optimizes memory retrieval, tool usage, and response style to tailor interactions.

• FABLE enables personalization of frozen, proprietary LLM agents without retraining the base model.
• It uses online learning from scalar feedback to adapt execution policies per user.
• The approach factorizes decisions into memory, information acquisition, and style layers.
• This method avoids the high cost of fine-tuning while improving user-specific relevance.

CrystalMem addresses memory hysteresis in self-evolving LLM agents, where capability drops after quota squeezes due to irreversible deletion or one-way compression. The paper proves that policies retaining only entries have a residual-deficit floor, then proposes a sidecar that demotes entries across four fidelity states using a crystallization-energy schedule. This approach orders demotions by advantage weighting to preserve essential knowledge while managing cloud resource constraints.

• Memory hysteresis occurs when agent capability fails to recover after budget cuts due to structural data loss.
• Standard keep-or-drop policies create a permanent performance deficit floor.
• CrystalMem uses a four-state fidelity system to manage memory elasticity without losing core knowledge.
• Demotion decisions are driven by an advantage-weighted crystallization-energy schedule.
• Acts as a sidecar, allowing existing LLM agents to adopt elastic memory management.

Self-improving LLM agents that rely on external memory for learning face a critical failure mode called the Echo Gap. Because ground-truth labels are unavailable in deployment, agents use LLM assessments as proxy rewards for stored episodes. This substitution leads to incorrect episodes receiving inflated scores, causing the system to preferentially reuse flawed experiences.

• Non-parametric self-improving agents are vulnerable to reward inflation in memory systems.
• Proxy rewards from LLM assessments can mislead policy improvement steps.
• The Echo Gap causes agents to reinforce incorrect behaviors over time.
• Score production mechanisms must be audited to prevent bias in retrieval.

Researchers introduce Bazaar, a dynamic sealed-bid benchmark designed to evaluate how well LLM agents price in real-world market conditions with hidden preferences and shifting demand. Unlike static tests, this framework uses closed-form customer utilities to allow for exact performance evaluation despite the environment's complexity. The study assesses 11 frontier models from four major providers, finding that top-tier agents like Gemini 3.1 Pro show competence in customer acquisition.

• Bazaar benchmarks LLMs in dynamic, multi-attribute auctions rather than static scenarios.
• Closed-form utilities enable exact evaluation of agent pricing performance.
• Tests reveal which frontier models can handle real-time market adaptation.
• Gemini 3.1 Pro leads in customer acquisition among the 11 evaluated agents.

RagTester is an automated framework designed to test the reliability of Retrieval-Augmented Generation systems by evaluating the interaction between generative models, embeddings, and retrieval mechanisms. It generates specific test inputs, documents, and expected outputs, then uses an LLM as a judge to evaluate the system's answers. The approach specifically targets complex passages, unsupported queries, and document coverage to ensure comprehensive validation.

• Automates the creation of test inputs, documents, and ground truth for RAG pipelines.
• Uses an LLM-as-a-judge to evaluate generated answers against expected outputs.
• Targets edge cases like unsupported queries and complex document passages.
• Validated across 24 configurations of eight LLMs and six embedding models.

AWS Transform continuous modernization is now generally available, allowing engineering teams to analyze and fix technical debt across large-scale codebases. Users can integrate with GitHub, GitLab, and Bitbucket to run scheduled or on-demand analyses covering security, agentic readiness, and custom criteria. The service automatically generates branches and pull requests for findings that have associated remediation strategies.

• Integrate GitHub, GitLab, or Bitbucket to scan for technical debt and security issues.
• Automate remediation by creating branches and pull requests for fixable findings.
• Prioritize work across technical debt, agentic readiness, and custom criteria.
• Run analyses on demand or via recurring schedules directly from the web app.

A new study characterizes AI coding agent workloads using 761M LLM calls from 3.2M GitHub Copilot users. The analysis shows sessions consist of sparse user turns that trigger autonomous agent loops with frequent tool execution. This structure results in high KV cache hit rates within turns but significant drops across turn boundaries.

• Agentic coding differs fundamentally from chatbots due to interleaved LLM inference and tool execution loops.
• KV cache hit rates average 90% within a single turn but drop to 55% across turn boundaries.
• Sparse user-initiated turns unfold into complex autonomous agent loops requiring distinct system optimizations.
• Production-scale data reveals unique workload properties with direct implications for inference infrastructure.

Nova is an automated end-to-end Just-In-Time compiler built on MLIR to bridge the gap between flexible tensor frameworks and physical hardware constraints. It captures eager executions to provide whole-graph visibility, enabling operation fusion, memory hierarchy optimization, and register-level tuning. This approach aims to maximize hardware utilization by offering granular control that eager execution models inherently lack.

• Captures eager executions to enable whole-graph optimization unavailable in standard frameworks.
• Fuses operations across boundaries to reduce overhead and improve throughput.
• Tunes execution down to the register level for maximum hardware utilization.
• Uses MLIR to provide granular control over hardware mapping and memory hierarchies.

AWS has enabled Amazon S3 Vectors in the European Sovereign Cloud (Germany) region. This service provides purpose-built vector storage for AI agents, RAG, and semantic search at billion-vector scale. It offers S3-level elasticity and durability with dedicated APIs, requiring no infrastructure provisioning.

• S3 Vectors is now live in the AWS European Sovereign Cloud (Germany).
• Supports billion-vector scale for AI agents, RAG, and semantic search.
• Provides S3-grade durability with dedicated APIs and no provisioning.
• Enables sovereign data residency for vector workloads in Germany.

Amazon EMR on EC2 now supports interactive Apache Spark sessions via Spark Connect, enabling data engineers to develop and debug applications directly from SageMaker Unified Studio, Jupyter, and VS Code. This feature introduces a persistent Spark context that spans multiple cells and scripts, allowing users to blend local Python execution with remote Spark operations. The underlying client-server architecture decouples the application client from the Spark driver, ensuring teams can retain their preferred local tooling while leveraging managed EMR clusters.

• Interactive Spark sessions persist across cells and scripts for seamless debugging.
• Client-server architecture decouples local IDEs from remote Spark drivers.
• Native integration with SageMaker Unified Studio, Jupyter, and VS Code.
• Active and completed sessions are monitorable via the EMR console.

Amazon RDS for SQL Server now supports Bring Your Own Media in ten additional commercial regions, including parts of Asia Pacific, Europe, and Mexico. This feature allows organizations to reuse existing Microsoft SQL Server licenses with active Software Assurance via the License Mobility program. Support covers SQL Server versions 2019, 2022, and 2025, with license tracking integrated into AWS License Manager for compliance.

• BYOM is now available in Asia Pacific, Europe, and Mexico regions for RDS SQL Server.
• Enables license reuse for SQL Server 2019, 2022, and 2025 with active Software Assurance.
• Integrated with AWS License Manager to track usage and ensure compliance.
• Expands flexibility for existing Microsoft customers migrating to managed AWS databases.

Amazon RDS for SQL Server now supports publishing native SQL Server Audit logs directly to CloudWatch Logs, in addition to the existing S3 option. This allows DBAs to perform real-time analysis of database engine events using CloudWatch metrics and alarms. When both S3 and CloudWatch are enabled, the system ensures logs are successfully uploaded to both destinations before marking the publication as completed.

• Publish SQL Server Audit logs to CloudWatch for real-time monitoring and alerting.
• Supports dual publication to S3 and CloudWatch with completion validation.
• Use CloudWatch Logs Insights to query audit data alongside other AWS logs.
• Maintain compliance by retaining logs on the DB instance via retention settings.

Fleet impact: For RDS SQL Server, this enables real-time alerting on sensitive database events via CloudWatch Alarms, reducing reliance on batch S3 processing. Ensure your CloudWatch Logs retention policies align with compliance requirements and monitor log ingestion costs if audit volume is high.

Amazon Aurora DSQL now supports multi-Region clusters in Europe (Stockholm), Europe (Spain), Asia Pacific (Mumbai), and Asia Pacific (Singapore). This serverless distributed SQL database offers active-active high availability and multi-Region strong consistency. Each cluster provides writable endpoints in both peered regions, ensuring a single logical database remains available during regional outages.

• Aurora DSQL multi-Region support now covers Stockholm, Spain, Mumbai, and Singapore
• Active-active architecture ensures strong consistency across peered regions
• Writable endpoints in both regions maintain availability during regional failures
• Expands the global footprint beyond US and existing APAC/Canada regions

Amazon RDS for Oracle has extended Reserved Instance pricing to include the new R8i and M8i instance families, offering up to 53% savings over On-Demand rates for 1- and 3-year terms. These instances utilize custom Intel Xeon 6 processors exclusive to AWS, delivering up to 2.5x the memory bandwidth and 15% better price-performance compared to previous Intel generations. The reservation benefits are applicable to both Single-AZ and Multi-AZ configurations, allowing flexibility to switch between them without losing savings.

• R8i/M8i RIs save up to 53% vs On-Demand for 1- or 3-year terms.
• Custom Intel Xeon 6 CPUs provide 2.5x memory bandwidth over prior gen.
• RI credits apply to both Single-AZ and Multi-AZ for the same class.
• Ideal for stable workloads needing high memory bandwidth performance.

Firecrawl released pdf-inspector, a Rust library that classifies PDFs as text-based, scanned, or mixed in under 50ms. It extracts text with position awareness and converts content to Markdown without invoking expensive OCR services. The tool provides bindings for Python, Node.js, and WebAssembly to handle local processing efficiently.

• Skip OCR costs for ~54% of text-based PDFs using smart classification routing.
• Get position-aware text extraction and Markdown conversion in under 200ms.
• Use Python, Node.js, or WebAssembly bindings for flexible integration.
• Detects scanned vs text-based types with confidence scores per page.