PingCAP, the company behind the TiDB database, confirmed unauthorized access to internal systems and theft of source code and customer-related data. Extortion group FulcrumSec claims 10.6 TB was taken, but these claims are unverified, and PingCAP says it has no evidence that data stored in TiDB was affected.
— via InfoQ, Renato Losio
An analysis of 78,612 real-world SBOM files found that 52.9% declare no dependency edges and 8.8% leave most components isolated, with only 38.3% forming well-connected graphs. The authors argue that treating 'no path' as 'unreachable' is unsound, and an explicit 'unknown' level restored KEV recall from 0.600 to 0.950 in a vulnerability-prioritization system.
— via arXiv Software Engineering, Artur Zi\k{e}ba-Kozarzewski
CSIRO built Serverless Beacon, a GA4GH Beacon implementation on AWS that queries genomic variants where they sit in S3 without centralizing the data. In testing on a 1.1 GB chromosome 1 VCF file, a 10,000 base pair query averaged 1.52 seconds at an estimated USD 0.00013.
— via InfoQ, Leela Kumili
The paper introduces AIRO, a framework for monitoring and responding to threats in Industrial IoT environments. It links Kafka streaming, ML inference and automated mitigations such as IP blocking, and its best models (Gradient Boosting, LightGBM) reach F1-scores up to 0.9229 on four purpose-built datasets.
— via arXiv Cryptography Security, Raima Deb, Pranav M Pawar, Mithun Mukherjee, Raja Muthalagu, Mian Guo
A study of Lightning Network gossip data infers node release lines and finds operators patch on release and platform schedules rather than in response to security disclosures. Half of LND nodes upgrade within 88 days, and 71% of exposed LND nodes still ran an affected release two months after an August 2026 disclosure.
— via arXiv Cryptography Security, Saurabh Kumar
A paper compares two deployments of a multi-tenant .NET core-banking platform: nine separate AWS ECS Fargate tasks versus a consolidated edition with a single API host. In emulation, the consolidated edition showed 2.3 times the read throughput, under half the memory and 11-24% lower estimated compute cost, but a larger failure domain. Results come from an emulated deployment, not production AWS.
— via arXiv Software Engineering, Tomisin Fagbola
The article argues that multi-cloud costs are driven mainly by architecture rather than provider rates, because cross-cloud data egress accumulates from service placement and communication patterns. In the controlled experiment, near-source processing saved over $2000 per month versus $825 from a better provider rate.
— via InfoQ, Vivek Arora
Docker has published a BAND AI kit for Docker Sandboxes so an isolated agent can join BAND rooms and work with humans and other agents. A walkthrough shows a local Repo Doctor agent repairing a failing Python test, with outbound-only connectivity and host-held credentials.
— via Docker Blog, Jennifer Kohl
The article compares two observability patterns for AI agents on Amazon EKS: a decentralized one where each agent sends traces straight to Langfuse, and a centralized one routing telemetry through a shared OpenTelemetry Collector. Both use Bifrost as an LLM gateway to Amazon Bedrock, with trace context holding traces together.
— via AWS Containers, Hari Muthusamy
Cilium 1.20 can now handle plain TCP and UDP traffic, such as databases and DNS, through Gateway API, so teams no longer need LoadBalancer or NodePort Services for them. A new setting, bpf.datapathMode=auto, also lets mixed-kernel clusters use netkit where supported and fall back to veth elsewhere.
— via InfoQ, Matt Saunders
GitHub's Actions Runner Controller 0.15.0 cuts down on Kubernetes API traffic by switching controller updates to patch requests and moving runner status aggregation into metrics. It also adds configurable reconciliation concurrency and listener rate limits, which the release notes caution can raise load if teams turn them up without testing.
— via InfoQ, Craig Risi
Buildkite's MCP server now lets AI coding agents check pipeline YAML against the official schema without uploading it or starting a build, and compare a failed build with the last good one on the same branch. It also reversed a default after its own evaluations showed that making failure summaries skip never-run jobs sent agents back for extra tool calls.
— via InfoQ, Matt Saunders
AWS has put an AI agent in preview that scans your accounts against Well-Architected best practices and hands back prioritized fixes, with CLI commands and IaC changes attached to each finding. It runs only in three US regions and requires a Business Support+ plan or higher, and AWS itself warns its recommendations may contain errors or be incomplete.
— via InfoQ, Renato Losio
On October 24, AWS will cut its European Sovereign Cloud off from the AWS Global Network for several hours, with traffic falling back to the public internet. AWS says neither disconnected system is operation-critical, and critics note a test of a few hours says nothing about the US parent's legal reach.
— via InfoQ, Steef-Jan Wiggers
