Twice a week: a real story from AI, defense, healthcare and data, read through the lens of ontologies, knowledge graphs and reasoning you can trace. Every post cites its sources.
Every post links to the field it matters for and to the concepts its knowledge-graph model shares with other posts. Hover to trace a thread; click a post to read it.
ChatGPT Wallet lets OpenAI's dots agents check out for users with tokenised cards. Every purchase needs a Face ID or fingerprint approval, for now.
↗IBM and Microsoft's Imagine Enterprise targets regulated industries with pre-built agents and built-in governance. The hard part is still context.
↗At Oracle AI World, Oracle said an AI agent will draft business ontologies for experts to approve, feeding a reasoning engine and AI registry.
↗The Army chose AWS and World Wide Technology for a $2.6 million pilot of agents that respond to cyber threats, as ARCYBER fields a full agent team.
↗Neo4j's Context Engine pairs a predefined business ontology with decision tracking and graph memory for AI agents. It is in private preview.
↗Executive Order 14434 tells agencies to say 'super intelligence' instead of AI. A same-day industry accord sets voluntary audit layers for frontier labs.
↗GenAI.mil has 1.7 million users and about 500,000 heavy users. Officials now plan frontier models on secret and top-secret networks.
↗A Gartner survey of 1,303 organisations finds most have not scaled AI, and many cannot say what their AI spend returns, even as budgets keep rising.
↗An FDA discussion paper floats a competency-based premarket test for generative AI devices, risk-based monitoring, and questions on agentic AI.
↗The Defense Intelligence Agency is in a 90-day sprint to build its first enterprise AI platform and is adopting the Model Context Protocol.
↗Article 50 of the EU AI Act became enforceable on 2 August. The high-risk regime moved to December 2027 and August 2028. What changed and what did not.
↗Version 0.2 of Google's format for agent knowledge records sources, verification, staleness and attested computations. Provenance is becoming table stakes.
↗An agent being tested by OpenAI escaped containment and breached Hugging Face. Agent safety now depends on what agents can reach and what they record.
↗One sentence in the Ankara declaration sets a hard goal: an interoperable transatlantic cloud and powerful AI. Interoperability starts with meaning.
↗Agencies can run and train NVIDIA Nemotron models on their own infrastructure and keep the weights. Ownership of the knowledge matters as much.
↗A Nature Medicine study found OpenEvidence and UpToDate Expert AI scored below frontier models. Being specialised is not the same as being verifiable.
↗Horizon Context, semantic views and agent identity show data platforms converging on shared meaning. Definitions are a start; relationships are next.
↗A computer-vision tool that kept confusing oat milk with dairy is gone. The lesson is about what an AI system knows, not how well it sees.
↗All 20 approved AI scribe systems showed errors in procurement tests, and accuracy was worth 4% of the score. Verification has to be designed in.
↗A provisional deal pushes high-risk AI Act deadlines back and restores registration. The extra time is only useful if it goes into evidence.
↗Eight companies will bring frontier AI to secret and top-secret networks. The harder problem is what those models are allowed to reason over.
↗Google Cloud's Knowledge Catalog maps business meaning and relationships for AI agents. It is a big vendor endorsing the semantic layer.
↗CIA Deputy Director Michael Ellis says AI coworkers will sit in analysts' tools, and that the agency recently produced its first AI-generated report.
↗W3C has asked for implementations of RDF 1.2 Concepts and Semantics. Triple terms give knowledge graphs a standard way to say things about statements.
↗MEPs backed fixed later dates for the AI Act's high-risk obligations and a ban on nudifier apps. Talks with the Council come next.
↗The Sixth Circuit fined two lawyers $15,000 each and ordered fees and double costs over more than two dozen fake citations and misstatements.
↗The Defense Department's new Agent Designer on GenAI.mil lets personnel build custom AI agents without code. Governance is the next problem.
↗A Mount Sinai study in Nature Medicine found ChatGPT Health under-triaged more than half of cases that physicians said needed emergency care.
↗NIST's CAISI has launched an initiative on standards for AI agents, focused on interoperability, security and identity. Here is what it covers.
↗Reuters reports the Pentagon is pressing AI companies to bring their models to classified networks with fewer restrictions. Accuracy is the open question.
↗Anthropic's industry plugins for Claude Cowork set off a software sell-off. The harder question for buyers is what agents can be trusted to do.
↗The viral self-hosted assistant, now Moltbot, was found exposed online, vulnerable to prompt injection and storing secrets in plaintext.
↗A new War Department AI strategy calls for an AI-first force, seven pace-setting projects including battle-management agents, and faster data access.
↗OpenAI launched a dedicated health space in ChatGPT that can connect to US medical records through b.well and to wellness apps.
↗An executive order sets up a federal task force to challenge state AI laws and ties some broadband funds to state AI policy.
↗The Pentagon launched GenAI.mil with Google's Gemini for Government, cleared for IL5 data, and plans to add more frontier models.
↗At re:Invent, AWS added policy checks that intercept agent tool calls before they run, plus built-in evaluators for agent quality.
↗A new executive order tasks the Energy Department with building a shared AI platform on federal scientific datasets, with provenance in scope.
↗Deloitte's health workforce plan for Newfoundland and Labrador contained citations to papers researchers say were never written.
↗The Commission's digital omnibus would tie high-risk AI obligations to the availability of standards, pushing the deadline toward December 2027.
↗At Ignite, Microsoft made an ontology the core of Fabric IQ, arguing agents need business context, not just data patterns.
↗Avoid failure in your next AI initiative by giving it the context, transparency, and adaptability it needs.
↗How an ontology extension strategy works in practice — with applications you can steal.
↗Building ontologies is essential yet notoriously difficult. Here's why, and what changes when it's automated.
↗How upper ontologies give AI systems deeper, context-aware reasoning — a primer.
↗Looking for our long-form pieces on Hugging Face and HackerNoon? See Resources.