AI News Archive: June 10, 2026 — Part 9
Sourced from 500+ daily AI sources, scored by relevance.
- SCANOSS Establishes US Presence to Meet Growing Enterprise Demand for AI-Speed Open Source Compliance
SCANOSS Establishes US Presence to Meet Growing Enterprise Demand for AI-Speed Open Source Compliance azcentral.com and The Arizona Republic
- The UK is running on fumes as data center build-outs can’t keep pace with demand
The UK is running on fumes as data center build-outs can’t keep pace with demand IT Pro
- Eufy Expands Face-Reading Smart Lock Line, Specializing in On-Device Processing
Eufy's FamiLock line is now larger, with more affordable models that limit AI processing to help protect your privacy. Here's how that works.
Score: 45🌐 MovesJun 10, 2026https://www.cnet.com/home/security/eufy-expands-face-reading-smart-lock-line-on-device-processing/ - Nashville Schools Use AI to Redesign District Org Chart
With the help from Vanderbilt University, staff developed a custom AI system that could assess, analyze and flag patterns and inconsistencies in job descriptions for roles with the same title.
Score: 45🌐 MovesJun 10, 2026https://www.govtech.com/education/k-12/nashville-schools-use-ai-to-redesign-district-org-chart - AWS’ powerful Graviton5 CPU makes its debut in new M9g and M9gd cloud instances
Amazon Web Services Inc.’s next-generation custom silicon is finally being made accessible to customers for the first time with the launch of the Elastic Compute Cloud M9g and M9gd instances. They’re powered by the all-new Graviton5 central processing unit, which is designed to deliver superior compute performance for diverse cloud workloads, including artificial intelligence applications. […] The post AWS’ powerful Graviton5 CPU makes its debut in new M9g and M9gd cloud instances appeared first on SiliconANGLE .
Score: 45🌐 MovesJun 10, 2026https://siliconangle.com/2026/06/10/aws-powerful-graviton5-cpu-makes-debut-new-m9g-m9gd-cloud-instances/ - Five ways York University is shaping the future of AI and digital health research
Five ways York University is shaping the future of AI and digital health research Toronto Star
- LLM Routing: From Strategy Selection to Production Architecture
Learn how LLM routing improves accuracy, latency, and cost with per request model selection. Optimize pipelines, and use the right tool every time.
- Some Spotify users are convinced the platform is creating AI artist profiles and bios, and I think these conspiracy theories are getting out of hand — and there are other music fans who agree
Spotify is facing more scrutiny over AI slop, but the accusations are getting out of hand.
- Building trust in enterprise AI: Together AI earns ISO 27001:2022 certification
Together AI has earned ISO 27001:2022 certification, validating our commitment to enterprise-grade security for production AI workloads.
- Will AI take jobs? Vijay Kedia points to refrigerators, electricity and highways for answers
In a post on X on 9 June, Vijay Kedia addresses the common concern that AI will take jobs. The ace investor says history suggests a different possibility. Refrigerators, electricity, highways, and the Internet created opportunities for entire ecosystems, and AI may follow the same path.
- How autonomous AI is transforming the future of enterprise innovation
Artificial intelligence is evolving beyond task-based automation toward autonomous systems capable of reasoning, planning, and executing complex workflows with minimal human intervention. As businesses prioritise measurable outcomes and real-world impact, Agentic AI is emerging as a key driver of enterprise transformation. Recognition platforms such as the ET Most Innovative AI Product Awards 2026 are spotlighting breakthrough innovations that are redefining the future of intelligent systems.
- AI Data Firm DDN Eyeing a Fresh Funding Round by End of Year
DDN, a data hardware and software provider that counts Nvidia Corp. as a partner and Alphabet Inc.’s Google and Salesforce Inc. as customers, is looking to raise additional funding by the end of the year, Chief Executive Officer Alex Bouzari said.
Score: 45💰 MoneyJun 10, 2026https://www.bloomberg.com/news/articles/2026-06-10/ai-data-firm-ddn-eyeing-a-fresh-funding-round-by-end-of-year - Why This Large District Used AI to Simplify Its Org Chart
The school system's organizational chart evolved organically for decades. Can AI bring cohesion?
Score: 45🌐 MovesJun 10, 2026https://www.edweek.org/technology/why-this-large-district-used-ai-to-simplify-its-org-chart/2026/06 - Gulf prepares new data centre investment
Gulf Development, Thailand's largest energy company by market value and a major telecom operator, has announced plans to invest in new data centres with a combined capacity of 100 megawatts this year, underscoring its ambition to become a regional leader in digital infrastructure and cloud services.
Score: 44🌐 MovesJun 10, 2026https://www.bangkokpost.com/business/general/3269153/gulf-prepares-new-data-centre-investment - CDEvents Simplifies AI-Ready Developer Platforms
CDEvents Simplifies AI-Ready Developer Platforms DevOps.com
- Exclusive: HyperNorm AI Nets $2.2 Mn To Build Wealth Advisory Intelligence Platform
Wealth management-focused AI startup HyperNorm AI has raised $2.2 Mn (about ₹19.5 Cr) in its seed funding round from Capital…
Score: 43💰 MoneyJun 10, 2026https://inc42.com/buzz/exclusive-hypernorm-ai-nets-2-2-mn-to-build-wealth-advisory-intelligence-platform/ - Caterpillar Holds Key Level After Plowing Into Data Center Market
Caterpillar stock is above its 50-day moving average on Wednesday after tight weekly closes with an entry at 931.35. The post Caterpillar Holds Key Level After Plowing Into Data Center Market appeared first on Investor's Business Daily .
- Channel Factory adds AI Slop Detection technology to its Proprietary AI content Classification System
Channel Factory has announced a new technology addition to its Proprietary AI tool – AI Slop Detection. This feature is in addition to contextual targeting technologies, which Channel Factory is providing in its Proprietary AI tool. The company is introducing a proprietary tiered AI content classification system that not only aligns with industry frameworks but goes significantly further […] The post Channel Factory adds AI Slop Detection technology to its Proprietary AI content Classification System appeared first on CXOToday.com .
- Beyond extract_text: The Two Layers of a PDF That Drive RAG Quality
Enterprise Document Intelligence [Vol.1 #5A] - Document signals (metadata, native TOC, source software) and page-level content (text vs scans, tables, images, columns, page profile) The post Beyond extract_text: The Two Layers of a PDF That Drive RAG Quality appeared first on Towards Data Science .
Score: 42🌐 MovesJun 10, 2026https://towardsdatascience.com/beyond-extract_text-the-two-layers-of-a-pdf-that-drive-rag-quality/ - Vu Technologies built over 100 studios. Now it’s focused on AI 'experiences'
Vu Technologies is transforming from a virtual production company into something CEO Tim Moore said is "much more high-tech than where we first started."
Score: 42🌐 MovesJun 10, 2026https://www.bizjournals.com/tampabay/news/2026/06/10/tim-moore-vu-technologies.html?ana=brss_6150 - Multimodal Browser AI with Transformers.js for Images and Speech
Most browser AI tutorials cover text because it is a natural starting point, but the applications people actually want to build are rarely text-only.
Score: 42🌐 MovesJun 10, 2026https://machinelearningmastery.com/multimodal-browser-ai-with-transformers-js-for-images-and-speech/ - Axitos.ai Launches Traditional Book Publishing Model, Opens Direct Submissions to Authors Across All Genres
Axitos.ai Launches Traditional Book Publishing Model, Opens Direct Submissions to Authors Across All Genres USA Today
- AI start-up Plaud to invest $10 million in Singapore as it expands Asia-Pacific operations
AI start-up Plaud to invest $10 million in Singapore as it expands Asia-Pacific operations The Straits Times
Score: 41🌐 MovesJun 10, 2026https://www.straitstimes.com/tech/plaud-to-invest-10-million-in-singapore-as-it-expands-asia-pacific-operations - DocLang aims to make documents readable by AI, not humans
AIs struggle to understand documents designed for humans; the DocLang working group seeks to flip that imbalance with its specification for machine-readable business documents “built from the ground up for LLM tokenizers.” The working group, founded by IBM, Nvidia, and Red Hat and hosted by the Linux Foundation’s LF AI & Data project, aims to create an open, universal, AI-native document format designed to improve how enterprises prepare, exchange, and govern document data for AI systems. ABBYY and Human Signal will also be involved in its development, and other contributors are welcome. “Enterprises today work across a fragmented landscape of document formats, including PDFs, JPEGs, and other file types built primarily for human consumption rather than AI interpretation,” the group said in its launch announcement . “This disconnect can introduce complexity, raise costs, and reduce reliability when extracting meaning from business documents,” as organizations increasingly rely on generative AI and agentic systems, it said. Mark Collier , executive director of LF AI & Data, said the goal of the DocLang Specification Working Group is to “develop a vendor-neutral, interoperable standard that helps organizations prepare document data for AI more reliably, transparently, and at scale.” DocLang defines a structured, machine-readable format for documents of any type, like JSON for data, that any tool can implement and any pipeline can consume. It builds on DocLing , a document processing toolkit hosted by LF AI & Data that can transform human-readable PDFs, word processor documents or spreadsheets into structured data. Standards must evolve for AI Something like DocLang is needed, said independent technology analyst Carmi Levy . “Existing document standards have done an admirable job allowing global stakeholders to confidently collaborate for decades, but it’s becoming increasingly clear that they are in desperate need of an update as AI reshapes the rules around how work gets done,” he explained. Largely static document types, he said, “can be somewhat limiting when AI is redefining the very word, ‘document.’ In many ways. AI-age documents are far more iterative and dynamic than what they once were, and the definitions need to evolve with the times. The documents we currently live with simply weren’t designed for the AI age.” Within that context, Levy said, “DocLang represents an early, best hope of achieving some kind of foundational baseline for document standards, one that will hopefully allow more intelligent, more efficient, lower-risk workflows than is currently the case.” Taking an open-source, vendor-agnostic approach to the process ensures the collective will take precedence over the needs of specific vendors, he said, adding, “earlier standards-setting efforts around networking, documentation, the web, and the cloud powered the free-flowing digital landscape that defines modern life.” An AI-centric documentation standard will carry that reality into the next generation of technology, said Levy. A question of governance The entire concept of LLMs, Jason Andersen , principal analyst at Moor Insights & Strategy said, “involves using natural human languages. The computer is supposed to understand us without us changing our syntax or language. Forcing a syntax on users is exactly what we have today with SEO and more advanced programming languages.” With something like DocLang, where the standard can be applied to content ingestion, he said, “I would be OK with that being automated, which seems to be the intent. The use case I envision is that when I upload a document to an agent, a skill can be run to preprocess the document into the DocLang standard format, saving tokens.” That makes sense, he said, adding that he thinks it’s good “if it can help generate outputs, like a visualization, that can be shared outside an AI tool. On that front, that is also why I am liking Web MCP, since you are just adding some code to the page, like CSS or JavaScript, and the consumer, in this case, an AI browser or skill, is better equipped to handle the site.” The point, he said, is, “these standards need to preserve the fact that humans can still do what they want, and do not need to know any coding to be proficient. In terms of governance, I am not sure if it matters.” But one analyst did foresee governance problems arising from DocLang’s use. Yaz Palanichamy , senior research analyst at Info-Tech Research Group, said DocLang adoption will require organizations to implement and review controls in order to scale its use accountably and securely.
Score: 41🌐 MovesJun 10, 2026https://www.cio.com/article/4183187/doclang-aims-to-make-documents-readable-by-ai-not-humans.html - Meet the 2026 Technology Pioneers shaping AI, energy, space and the next frontier
The Forum has announced its latest cohort of Technology Pioneers. Here's what they tell us about innovation, entrepreneurship and frontier technology.
Score: 41🌐 MovesJun 10, 2026https://www.weforum.org/stories/2026/06/meet-the-2026-technology-pioneers/ - AI model “hears” whale calls in seismic data
AI model “hears” whale calls in seismic data EurekAlert!
- The Man Behind Anthropic's Claude Code Hasn't Written a Line of Code in 8 Months — Here's What He Does Instead
The Man Behind Anthropic's Claude Code Hasn't Written a Line of Code in 8 Months — Here's What He Does Instead entrepreneur.com
- AI skills: The next layer of marketing automation by Optmyzr
Learn how to install, deploy, and brand AI skills to turn generic chatbots into your agency’s custom, scalable operating system. The post AI skills: The next layer of marketing automation appeared first on MarTech .
- Why Frictionmaxxing Could Be the Secret Weapon Brands Need in the AI Era
Decades of consumer psychology said to remove friction. AI is forcing a different tack.
- Which AI models can you automate on Zapier? (Opus 4.8, Gemini 3.5 Flash, and more)
New AI models launch practically every week, and keeping up with which ones to use for specific workflows is a job in itself. Consider this article your living reference. At Zapier, we run every model through AutomationBench. It's our benchmark for testing how well models carry out multi-step workflows, not just static prompts. Below, I'll walk through every major AI provider available on Zapier, the models you can plug into your Zap workflows today, and what each one is best for based on Zapier
- Will your iPhone support Siri AI? The answer is complicated
With iOS 27, only certain iPhone models and other devices will support all Siri features. Here's how this works.
Score: 40🌐 MovesJun 10, 2026https://www.zdnet.com/article/will-your-iphone-support-siri-ai-the-answer-is-complicated/ - AI Boom Returns Data Storage Tycoon Peter Shu To The Ranks Of Taiwan’s Richest
Shares of Transcend Information more than tripled from a year ago, ushering chairman and CEO Peter Shu back onto the list after an eight-year gap.
- How memory tools can make AI models worse
New research suggests that AI memory systems can degrade model performance and encourage sycophantic tendencies.
Score: 40🌐 MovesJun 10, 2026https://techcrunch.com/2026/06/10/how-memory-tools-can-make-ai-models-worse/ - Soccer Fans, You’re Being Watched
From anti-drone tech to face recognition, 2026 World Cup stadiums in the US, Canada, and Mexico are subjecting fans to an array of surveillance tech. Here’s what you need to know.
- AI SQL Sentinel: A Database Performance Monitoring Tool Built on Operational Knowledge
Most database monitoring tools tell you something is wrong. The harder problem is explaining what is wrong, why it happened, and what to do about it — quickly enough to matter. AI SQL Sentinel is a performance monitoring system for SQL Server and PostgreSQL that automates that diagnostic step. It collects live database evidence, identifies known failure patterns, retrieves matching expert guidance, and produces a structured diagnosis with a root cause, a confidence score, and executable resolution steps. The investigation that would otherwise take 20 to 40 minutes of manual querying happens in under a minute. This article describes what the tool does, how it is structured, and the design decisions behind it. What It Produces Before describing the architecture, here is the actual output from a live run against a CDC log pressure incident: Type: cdc_log_pressure Job: CDC_LOG_SCAN Root Cause: CDC scan session stuck for 420s with 0 commands processed. Transaction log held open — WRITELOG cascading to all concurrent DML. Confidence: 87% Evidence: • scan_status=STUCK_ACTIVE, duration=420s, command_count=0 • avg_write_ms=18.3 on OpsInsights (threshold: 5ms) • WRITELOG waits on 4 concurrent sessions during job window Recommended Actions: 1. EXEC sys.sp_cdc_stop_job 'capture'; EXEC sys.sp_cdc_start_job 'capture'; 2. Tune sp_cdc_change_job @maxtrans=2000, @maxscans=20 3. Verify log disk throughput — avg_write_ms > 10ms indicates disk saturation Recurring: seen 3 times in last 30 days | Blocking co-fired 2 of 3 times Root cause is evidence-based, not generic. Confidence reflects actual certainty. Evidence cites measured values from the database. Recommended actions are specific, executable SQL. The recurring pattern and co-occurrence note come from stored incident history, not from inference. Why Traditional Monitoring Wasn’t Enough Traditional monitoring platforms excel at collecting metrics and generating alerts. They are intentionally conservative about drawing conclusions because diagnosing root cause requires context, history, and operational judgment that a generic alerting system cannot reliably supply. The result is a consistent pattern: the alert fires, a DBA connects, and the manual investigation begins — checking wait stats, examining blocking chains, reviewing execution plans, correlating timing across subsystems. The monitoring tool told you something was wrong. It did not tell you why. That investigation takes 20 to 40 minutes for a familiar failure pattern and longer for anything unusual. AI SQL Sentinel focuses on that diagnostic layer. Rather than asking an operator to manually correlate evidence, it attempts to assemble the evidence, retrieve relevant operational knowledge, and produce a structured diagnosis before a human begins the investigation. The monitoring alert becomes the starting point of a resolved incident rather than the starting point of a manual search. The distinction is not that traditional tools are inadequate — it is that they were designed for a different job. Metric collection and alerting are solved problems. Automated diagnosis from live evidence is not. How It Works — Six Stages Every run works through the same six stages in sequence, fully automatic. Database Evidence → Expert Reasoning → Actionable Diagnosis 1 · Collect — Gathers evidence without touching the monitored database. 2 · Detect — Identifies which failure patterns are active. 3 · Suppress — Filters noise before any AI call is made. 4 · Retrieve — Finds the most relevant expert guide from the knowledge base. 5 · Explain — Combines the evidence and guide into a structured diagnosis. 6 · Learn — Stores the outcome so future incidents on the same target have more context. Architecture: Three Databases, Two Processes The system uses three distinct database connections, each with a different role. Monitored SQL Server — evidence source only. The tool reads sys.dm_exec_requests, ring buffers, Query Store, job history from msdb, and related system views. It never writes to the monitored server. The login requires VIEW SERVER STATE and SQLAgentReaderRole in msdb — no elevated permissions, no schema changes. Monitored PostgreSQL — evidence source only, read via pg_monitor role. The tool reads pg_stat_activity, pg_stat_bgwriter, pg_replication_slots, pg_class, and related catalog views. PostgreSQL repository on Neon — the only database the tool writes to. All persistent storage lives here: incidents, alert quality records, job history ingested from SQL Server, 5-minute snapshots, and the runbook knowledge base with vector embeddings. Two processes run on a schedule: collect_snapshots.py runs every five minutes. It photographs the current state of the monitored database — blocking chains, deadlock events, memory queues, CPU ring buffer, CDC scan state, replication lag — and writes everything to the Neon repository. This enables historical replay and pattern learning. detect_issues.py runs on demand or on a schedule. It reads those snapshots or queries live views directly, runs all detectors, calls the AI model, and stores validated diagnoses. It can run against today's live data or replay any historical date using the --date flag. The Detection Layer: 25 Failure Patterns Every detector query fires only when a threshold is crossed. Zero rows returned means nothing to investigate. The AI is never called on normal activity. SQL Server — 13 detectors Job baseline comparison checks each job’s last 24 hours against its 30-day P50 median. It requires at least 10 baseline runs for statistical stability and fires when a job runs more than twice its normal time. Job trend detection uses linear regression on 14 days of run history to surface slow-burn deterioration before it crosses the 2× threshold. It fires when the trend is clearly upward but the job has not yet doubled — an early warning that arrives 2 to 5 days before the condition becomes an incident. Live blocking checks for active lock waits and blocked sessions in the current DMV snapshot. Deadlock detection parses the SQL Server system health extended events ring buffer for deadlock events in the last two hours. The ring buffer holds approximately 256 events and resets on server restart. The snapshot table in the repository provides durable storage that survives restarts. Memory pressure fires when sessions are queuing for memory grants (RESOURCE_SEMAPHORE wait type) or server RAM exceeds 92%. CPU pressure fires when average SQL Server CPU exceeds 80% across ring buffer samples. The AI prompt contains an explicit diagnostic rule for this: low CPU time combined with high elapsed time is memory starvation, not CPU saturation — a distinction that changes the fix path entirely. CDC log scan pressure checks CDC scan sessions for the stuck state — open more than five minutes with zero commands processed. Also monitors log write latency above 5ms. Completely silent on instances without CDC enabled. Long-running execution with severity scoring scores each active session on 11 compounding signals: elapsed time, being blocked by another session, blocking other sessions, holding an open transaction, critical wait types, high logical reads, write pressure, and CPU time. A backup running alone scores 1 and is ignored. A session running 87 minutes while blocking three others and holding an open transaction scores 9 and fires immediately. Multiple concurrent risk signals are required before a high-severity alert fires. Query intelligence watches for seven structural query problems: missing indexes with cardinality analysis, plan instability across Query Store executions, top resource-consuming queries by logical reads, unused high-cost indexes on write-heavy tables, stale statistics, implicit type conversions forcing table scans, and sampled statistics quality anomalies. The last signal fires when statistics were recently updated but the sample rate was too low to accurately represent the value distribution — auto-update alone does not fix it, and the resolution path differs from stale statistics. tempdb pressure watches four distinct failure modes: session spills over 500MB, version store growth over 10GB, allocation page contention from too few tempdb data files, and file capacity approaching 85%. Each has a different fix. All four cause every query to slow simultaneously with moderate CPU and no blocking — a signature that no other detector catches. Storage I/O saturation fires when data file read or write stall exceeds 20ms per IO, or when five or more sessions are waiting for pages to load from disk simultaneously. Hot page latch contention fires when three or more threads compete simultaneously for an exclusive in-memory page latch on the same user database B-tree page — the pattern caused by clustered indexes with sequentially increasing keys. The fix is at the schema level. Worker thread starvation requires two signals simultaneously: live tasks queued with no worker thread assigned, and non-zero work queue counts on online schedulers. Neither alone is sufficient. Both together confirms the server is failing to hand out workers — new requests queue, SQL Agent jobs fail to start, logins may time out. This condition can develop and cause an outage in 60 to 90 seconds. A session-duration detector with a 30-minute threshold would never catch it. PostgreSQL — 12 detectors The PostgreSQL path covers blocking via pg_blocking_pids(), deadlocks, memory pressure via bgwriter statistics, CPU via active session patterns, WAL replication lag including inactive slots retaining WAL on disk, long-running sessions, query performance via pg_stat_statements, I/O pressure via cache miss rate, table bloat from dead tuple accumulation, vacuum lag and XID wraparound risk, and a three-signal visibility horizon blocker for stale replication slot xmin, forgotten prepared transactions, and MultiXact ID exhaustion. Suppression: Keeping the Signal Clean Four rules run before the AI is called, ensuring the AI only runs on issues that are real, new, and worth investigating. Maintenance window — if a configured maintenance window is active, non-critical alerts are suppressed silently. The window is set in the environment file and supports midnight-crossing ranges. Known workload exceptions — jobs listed as expected-slow are permanently skipped. Nightly index rebuilds, end-of-month archival jobs, and similar predictable long-running work do not generate alerts. Duplicate Alert Prevention — if the same issue type fired for the same job in the last four hours, it is suppressed. The same active problem does not flood the team. Severity threshold — for the long-running session detector, only sessions scoring 3 or above proceed. Low-scoring sessions are logged to the alert quality table but do not reach the AI. Eight issue types bypass all four rules regardless: deadlocks, memory pressure, CPU pressure, tempdb pressure, I/O saturation, hot page latch contention, worker thread starvation, and the PostgreSQL visibility horizon blocker. Active failure conditions that can cause immediate data loss or outage are never suppressed. Every suppression decision is written to the alert quality table with its reason. The team can always query what was seen, what score it received, and why it was held back. Runbook Retrieval: Finding the Right Expert Guide Before the AI sees any evidence, it retrieves the most relevant troubleshooting guide from the knowledge base. The retrieval uses two gates. Hard category filter first. Each detector produces a fixed issue type string that maps to a runbook category. A SQL query restricts the search to only that category. A tempdb runbook cannot compete against a blocking runbook regardless of vocabulary. The category is determined by the detector, not by text similarity. Semantic similarity search second. Within the filtered category, the issue description is encoded into a 384-dimensional vector using all-MiniLM-L6-v2 and compared against pre-computed runbook embeddings using cosine distance via pgvector. Results below a 0.40 similarity score are dropped — an irrelevant runbook is worse than no runbook. The knowledge base contains 37 runbooks — 27 for SQL Server, 10 for PostgreSQL. Each contains four sections: a signal fingerprint written to match the vocabulary the detector produces, ranked root cause paths ordered by likelihood, investigation and resolution steps from least invasive to most disruptive, and an explicit section on what not to do. The wrong-path section exists because an AI model given plausible evidence will sometimes recommend the wrong fix confidently. Encoding the wrong answers prevents that. The runbooks cover every detector type — job slowdown triage, blocking and lock contention, uncommitted transaction blocking, ETL and read contention, all seven query intelligence signals, hot page latch contention, storage I/O saturation, worker thread starvation, plan regression, parameter sniffing, deadlock resolution, memory grant pressure, CPU saturation, CDC log scan lag, long-running session triage, tempdb pressure, and the full PostgreSQL coverage including visibility horizon blockers, table bloat, vacuum lag, WAL replication lag, memory and CPU pressure, query intelligence, I/O saturation, and temporary space pressure. The AI Prompt: Nine Sections The prompt the AI receives is assembled from nine sections, each with a specific job. Section 1 assigns the engine-specific persona — SQL Server DBA or PostgreSQL DBA. The PostgreSQL persona contains an explicit prohibition against suggesting SQL Server tools. Without this, the model will occasionally suggest SQL Server-specific commands for PostgreSQL incidents. Section 2 provides a ten-step reasoning framework the AI must work through in sequence: what changed, is blocking present and when did it start, is memory or CPU the constraint, is CDC holding the log, is another database consuming disproportionate resources, is a plan regression visible, is the pattern time-of-day related, what is the safest first action. Section 3 provides the incident summary — issue type, job name, current and baseline values. Section 4 handles temporal validation. Suspected causes must have occurred during the incident window. This prevents a specific wrong diagnosis: a job ran slow, a blocking chain appeared in a snapshot taken after the job completed, and the AI attributes the slowdown to that blocking. With the temporal rule, the AI identifies it as coincident, not causal. Section 5 provides query regression detail from Query Store when available — plan IDs before and after, plan count, execution counts. Section 6 provides blocking evidence sanitised and capped at eight sessions, with cross-database blocking explicitly flagged. Section 7 provides resource pressure evidence formatted specifically for the issue type that fired. The format for a tempdb incident differs from the format for a memory pressure incident. Section 8 provides server-level context — which database is consuming the most CPU, reads, and log write latency, with the monitored job’s database marked. Section 9 injects the retrieved runbook with its similarity score visible to the model. The score signals how heavily to weight the retrieved guidance versus the model’s own evidence reasoning. Historical patterns from stored incidents are also included: whether this issue has recurred in the last 30 days, whether it follows a time-of-day pattern, and which other issue types have co-fired alongside it in past incidents. If blocking has co-fired with deadlocks in 80% of past incidents on this target, the AI is told explicitly and this is treated as a strong signal when forming the root cause. The output is structured JSON — root cause, confidence score as a float between 0 and 1, evidence array citing observed values only, and ranked recommended actions. Schema validation runs before anything is stored. Malformed responses produce a safe fallback rather than a lost incident. Security Sanitisation is hardcoded on and cannot be disabled. Login names are hashed with SHA-256. Hostnames are replaced entirely. SQL text is truncated at the database level before it leaves the server — enough to identify the operation, not enough to expose literal values from WHERE clauses. Deadlock query text is capped at 80 characters. Five authentication modes are supported for SQL Server: standard SQL login, Windows integrated authentication, Azure Managed Identity, Azure AD password, and Kerberos. In cloud and domain-joined environments the SQL login credential can be eliminated entirely. The monitored SQL Server login requires VIEW SERVER STATE and SQLAgentReaderRole in msdb. No sysadmin. No write access. No schema creation on monitored systems. Alert Quality Measurement The system measures its own quality. The alert quality table records every firing decision and every suppression with its reason. Operators can label incidents as true positive, false positive, or noise using a CLI command. python detect_issues.py --quality-report The report shows suppression rate, mean AI confidence per issue type, detector activity over the selected period, repeat offenders, and precision once enough operator feedback has been recorded. Current State The system has been validated against reproduced failure scenarios — real patterns recreated using sample databases in a controlled environment. It has not yet been deployed against a live production database. What has been built: 25 detectors, 37 runbooks, 532 unit tests requiring zero database connections, read-only access throughout, no agents, no writes to monitored systems, configurable AI provider (Anthropic Claude or OpenAI, switched via a single environment variable). What comes next: a production pilot to measure detection accuracy and diagnostic quality against real workloads, and integration with Slack and email notification so incidents reach the team without requiring manual polling. Limitations and Open Questions The system has been validated against reproduced failure scenarios, but several areas remain intentionally unmeasured until a production pilot is completed. False positive rate is not yet available because no production incident data exists. The alert quality framework and operator feedback workflow have been implemented to measure this during the pilot. Runbook retrieval accuracy has been validated during development through category-gated retrieval and similarity thresholds, but has not yet been benchmarked against a representative production dataset. Runbook maintenance and drift remain a human-reviewed process. Incident history and AI diagnoses are retained to support future refinement, but no automated drift detection currently exists. Detector maintenance is intentionally lightweight, with thresholds externalised from detection logic, though future platform and DMV changes may require periodic updates. AI and runbook disagreements are preserved for review rather than automatically overridden. The runbook serves as guidance, while the final diagnosis remains evidence-driven. The production pilot will focus on answering these questions with real operational data. The tool does one thing: when a database issue occurs, it produces a diagnosis — root cause, evidence, confidence, and specific actions — before a DBA has had time to open a query window. How much of the subsequent investigation that eliminates depends on what the pilot reveals. Technical reference: the complete detector query documentation, DMV and system view matrix, prompt structure, JSON output schema, and runbook content are documented separately in the project repository. AI SQL Sentinel: A Database Performance Monitoring Tool Built on Operational Knowledge was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
- EHang Aircraft Sales Plunge as eVTOL Commercial Rollout Stalls
EHang Aircraft Sales Plunge as eVTOL Commercial Rollout Stalls Caixin Global
- Who Gets Hired In An AI-First World?
Ever since they have been around, technology companies hired for pedigree, technical depth, experience and execution speed. But, with AI…
- AI jobs apocalypse is overblown, says Starmer adviser
AI jobs apocalypse is overblown, says Starmer adviser The Telegraph
Score: 40🌐 MovesJun 10, 2026https://www.telegraph.co.uk/business/2026/06/10/ai-jobs-apocalypse-is-overblown-says-starmer-adviser/ - Anthropic leans into AI’s nascent slice-and-dice era
Clever financial engineering is allowing conservative, risk-averse investors to participate enthusiastically
- What smart people are saying about the 2 most controversial parts of Anthropic's new models
What smart people are saying about the 2 most controversial parts of Anthropic's new models Business Insider
Score: 40🌐 MovesJun 10, 2026https://www.businessinsider.com/what-smart-people-are-saying-about-anthropics-new-ai-limits-2026-6 - Chairman Of AI Chip-Testing Equipment Maker Is The Wealthiest Newcomer Among Taiwan’s Richest
The red-hot November IPO of AI chip-testing equipment maker Hon. Precision put chairman Hsieh Wen-Ta on the ranking of Taiwan’s richest for the first time with a $7.1 billion fortune.
- AppSquadz Earns AWS AI Services Competency: Generative AI Consulting Services, Strengthens Leadership in Enterprise AI and Cloud Innovation
AppSquadz has earned the AWS AI Services Competency: Generative AI Consulting Services, a prestigious recognition which reinforces the brand’s proven capability to design, deploy and scale secure, enterprise-grade ready-to-deploy generative AI solutions on Amazon Web Services (AWS). The achievement marks a significant milestone in the company’s AI transformation journey and validates its expertise in enabling […] The post AppSquadz Earns AWS AI Services Competency: Generative AI Consulting Services, Strengthens Leadership in Enterprise AI and Cloud Innovation appeared first on CXOToday.com .
- Meta launches program to train workers for data center jobs
Meta pledged to invest $115 million to train electricians, plumbers and other workers needed to operate data centers.
Score: 40🌐 MovesJun 10, 2026https://www.cbsnews.com/news/meta-data-center-workforce-academy-training/ - AI Search Visibility Tools For Professionals Announced Amid ChatGPT Usage Trends
AI Search Visibility Tools For Professionals Announced Amid ChatGPT Usage Trends USA Today
- Anthony Seldon: AI makes children workshy
Anthony Seldon: AI makes children workshy The Telegraph
Score: 38🌐 MovesJun 10, 2026https://www.telegraph.co.uk/news/2026/06/10/anthony-seldon-ai-makes-children-workshy/ - Wise alumni raise $2m pre-seed for AI tax startup
Wise alumni raise $2m pre-seed for AI tax startup
- An ab initio study and machine learning framework to capture the motional effects in solid-state NMR of lithium-ion conductors - ORA - Oxford University Research Archive - ORA
An ab initio study and machine learning framework to capture the motional effects in solid-state NMR of lithium-ion conductors - ORA - Oxford University Research Archive ORA - Oxford University Research Archive
- How SA’s SupaChat is helping businesses deploy AI agents so they never miss a lead
South African startup SupaChat Global enables businesses to deploy AI agents trained on their own products, pricing, brand voice, and customer service parameters, engaging customers across WhatsApp, web, Facebook and Instagram, in multiple languages, around the clock. Founded last year, SupaChat is the “deployment layer” for AI in Africa. “Big Tech builds the models and [...] The post How SA’s SupaChat is helping businesses deploy AI agents so they never miss a lead appeared first on Disrupt Africa .
- Thirty-five AI comedians walked into a workshop, and what happened next could reshape how machines learn humor
Workshopping, an iterative process in which creators share ideas, test what works and refine what doesn't through collective feedback, is at the heart of any writers group. This collaborative dynamic inspired George Mason University Ph.D. student Shiwei Hong to explore whether artificial intelligence (AI) could benefit from a similar approach.
Score: 38🌐 MovesJun 10, 2026https://techxplore.com/news/2026-06-ai-comedians-workshop-reshape-machines.html - US-DATA Expands AI Data Annotation Services
US-DATA Expands AI Data Annotation Services USA Today
Score: 37🌐 MovesJun 10, 2026https://www.usatoday.com/press-release/story/33076/us-data-expands-ai-data-annotation-services/