https://www.noze.it/en/insights/ noze articles and analysis on the EU AI Act, NIS2, GDPR, digital health and open-source software. en-GB Fri, 10 Jul 2026 00:00:00 GMT https://www.noze.it/en/insights/agentic-loop-3-security-governance/ https://www.noze.it/en/insights/agentic-loop-3-security-governance/ Fri, 10 Jul 2026 00:00:00 GMT To work, the loop reads untrusted content into the context and has tools to act: that is its attack surface. Part 3 of 3: prompt injection and the lethal trifecta, why security lives in the harness and not the prompt, observability, and governance in the loop (least privilege, human-in-the-loop, runtime policy, audit). AI Cybersecurity Governance https://www.noze.it/en/insights/harness-engineering/ https://www.noze.it/en/insights/harness-engineering/ Thu, 09 Jul 2026 00:00:00 GMT The model decides what to write; the harness governs when, where and how. Harness engineering is the discipline of building the environment around the model (instructions, state, verification, scope, lifecycle) so it produces reliable results. What OpenAI's Codex experiment (a million lines, none written by hand), Anthropic's ($9 vs $200) and the open source course on the topic teach. AI Cybersecurity R&D https://www.noze.it/en/insights/agentic-loop-2-context-and-patterns/ https://www.noze.it/en/insights/agentic-loop-2-context-and-patterns/ Wed, 08 Jul 2026 00:00:00 GMT The basic cycle is simple; making it last is not. Part 2 of 3: context engineering (truncation, compaction, externalized state, RAG, sub-agent offload), the patterns beyond reactive (ReAct, Plan-and-Execute, Reflexion, verification), multi-agent as loops of loops, robustness and the latency-and-cost bill. AI Cybersecurity R&D https://www.noze.it/en/insights/claude-j-space/ https://www.noze.it/en/insights/claude-j-space/ Tue, 07 Jul 2026 00:00:00 GMT On 6 July 2026 Anthropic published 'A global workspace in language models': inside Claude there is a small internal workspace, the J-space, that emerged on its own during training and holds the reportable, reusable concepts while about 90% of the computation runs automatically. How the J-lens works, why it matters for safety, and why they talk about access, not consciousness. AI Cybersecurity Governance https://www.noze.it/en/insights/agentic-loop-anatomy/ https://www.noze.it/en/insights/agentic-loop-anatomy/ Mon, 06 Jul 2026 00:00:00 GMT An LLM is stateless: an agent is what you get when you put it in a loop with a goal, tools, observations and a stop condition. Part 1 of 3: the minimal loop (with pseudocode and a diagram), the components, schema-typed tool calling and termination. AI Cybersecurity R&D https://www.noze.it/en/insights/fable-5-redeployment-cybersecurity/ https://www.noze.it/en/insights/fable-5-redeployment-cybersecurity/ Thu, 02 Jul 2026 00:00:00 GMT On 1 July 2026 Anthropic restored access to Claude Fable 5 (and Mythos 5 for US defenders only, via Project Glasswing) after export controls were lifted. New anti-jailbreak safeguards validated by CAISI, a severity framework for cyber-jailbreaks, and an Executive Order coordinating government and frontier labs. What it means for defenders. AI Cybersecurity Governance https://www.noze.it/en/insights/dwarfstar-4/ https://www.noze.it/en/insights/dwarfstar-4/ Tue, 30 Jun 2026 00:00:00 GMT antirez (Salvatore Sanfilippo, creator of Redis) has released DwarfStar 4 (ds4): a native, self-contained, open source (MIT) inference engine that runs DeepSeek V4 Flash, a quasi-frontier model, locally on 96-128 GB of RAM thanks to an asymmetric 2-bit quantization. What it is, how it works, and why it matters for AI sovereignty. AI Open Source Governance https://www.noze.it/en/insights/aleph-neuro-ultrasound-brain/ https://www.noze.it/en/insights/aleph-neuro-ultrasound-brain/ Sun, 28 Jun 2026 00:00:00 GMT On 24 June 2026 Aleph Neuro published what it claims is the first 3D image of the vasculature of a living human brain captured with ultrasound through the intact skull, using ultrasound localization microscopy (ULM), releasing the full pipeline and dataset as open source on GitHub. How it works, why it matters for digital health, and our take. Digital Health R&D Open Source https://www.noze.it/en/insights/gpt-5-6-sol/ https://www.noze.it/en/insights/gpt-5-6-sol/ Fri, 26 Jun 2026 00:00:00 GMT On 26 June 2026 OpenAI previewed GPT-5.6 Sol, Terra and Luna in a limited release: stronger capabilities in coding, biology and cybersecurity, but initial access restricted to a small group of partners agreed with the US government. Two weeks after the Fable 5 shutdown, frontier AI becomes gated technology. What it means for access and sovereignty in Europe. AI Governance Compliance https://www.noze.it/en/insights/sakana-fugu/ https://www.noze.it/en/insights/sakana-fugu/ Tue, 23 Jun 2026 00:00:00 GMT On 22 June 2026 Sakana AI unveiled Fugu, an orchestrator model that coordinates a pool of frontier LLMs behind a single API. What it is, how it works (TRINITY and Conductor), the reported benchmarks, the anti vendor lock-in thesis, and why it isn't available in the EU/EEA at launch due to GDPR. AI Governance Compliance https://www.noze.it/en/insights/glm-5-2-open-weight-ai-sovereignty/ https://www.noze.it/en/insights/glm-5-2-open-weight-ai-sovereignty/ Sat, 20 Jun 2026 00:00:00 GMT Z.ai releases GLM 5.2 with open weights (MIT) and a 1M-token context. According to Artificial Analysis it's the best open-weight model and fourth overall. What it means for organisations that want frontier AI they can control and self-host. AI Open Source Governance https://www.noze.it/en/insights/midjourney-medical/ https://www.noze.it/en/insights/midjourney-medical/ Thu, 18 Jun 2026 00:00:00 GMT On 18 June 2026 Midjourney announced Midjourney Medical: a scanner that reconstructs the whole body with ultrasound in 60 seconds, with AI for reconstruction and segmentation, and a network of 'spas' for mass preventive imaging. What was unveiled, the objections to expect from the clinical world, and why we think innovations like this, from large groups and small startups alike, will become ever more frequent. Digital Health AI R&D https://www.noze.it/en/insights/fable-5-shutdown-us-government-ai-sovereignty/ https://www.noze.it/en/insights/fable-5-shutdown-us-government-ai-sovereignty/ Sun, 14 Jun 2026 00:00:00 GMT On 12 June 2026 the US government ordered Claude Fable 5 and Mythos 5 suspended, and Anthropic disabled them worldwide within hours. What it teaches European organisations about dependency, business continuity and AI sovereignty. AI Governance Compliance https://www.noze.it/en/insights/decreti-attuativi-legge-ai/ https://www.noze.it/en/insights/decreti-attuativi-legge-ai/ Thu, 11 Jun 2026 00:00:00 GMT Italy's Council of Ministers gave preliminary approval to two decrees implementing Law 132/2025: AgID and ACN as national authorities, mandatory training across sectors, police biometrics, civil liability and the new Article 437-bis of the criminal code. What changes for companies, public bodies and professionals. AI Compliance Governance https://www.noze.it/en/insights/cyber-resilience-act-sbom/ https://www.noze.it/en/insights/cyber-resilience-act-sbom/ Tue, 09 Jun 2026 00:00:00 GMT The Cyber Resilience Act (Reg. EU 2024/2847) is moving forward: authorities from 11 June 2026, vulnerability reporting from September, SBOM and full obligations from December 2027. What changes, also for Open Source. Compliance Cybersecurity Open Source Governance https://www.noze.it/en/insights/gemma-nuova-release/ https://www.noze.it/en/insights/gemma-nuova-release/ Fri, 05 Jun 2026 00:00:00 GMT Google releases Gemma 4 under the Apache 2.0 licence, dropping the custom Gemma Terms of Use. A compliance decision before a product one, with the multimodal 12B variant running on a laptop. Open Source AI Compliance Governance https://www.noze.it/en/insights/ai-act-omnibus-adozione/ https://www.noze.it/en/insights/ai-act-omnibus-adozione/ Wed, 03 Jun 2026 00:00:00 GMT The Digital Omnibus on AI is heading toward formal adoption but is not law yet: until publication in the EU Official Journal the new dates are not binding and the August 2026 threshold formally stands. What companies should do now. AI Compliance Governance https://www.noze.it/en/insights/workflow-claude-code-tutorial/ https://www.noze.it/en/insights/workflow-claude-code-tutorial/ Mon, 01 Jun 2026 00:00:00 GMT Hands-on tutorial: build a multi-agent review workflow in Claude Code with the agent(), pipeline() and parallel() primitives, reproducing the find-adversarial verify-synthesise pattern of the bundled deep-research command. The feature is in research preview. AI Governance Open Source Compliance https://www.noze.it/en/insights/claude-opus-4-8-workflow/ https://www.noze.it/en/insights/claude-opus-4-8-workflow/ Fri, 29 May 2026 00:00:00 GMT In late May 2026 Anthropic released Claude Opus 4.8 and introduced Dynamic Workflows in Claude Code: scripts that orchestrate subagents at scale, in pipelines and in parallel, with validated structured output. The feature is in research preview. The angle: codified orchestration shifts the risk from the model to the process, and governance becomes the enabler. AI Governance Compliance https://www.noze.it/en/insights/nis2-pmi-cyberscan/ https://www.noze.it/en/insights/nis2-pmi-cyberscan/ Wed, 27 May 2026 00:00:00 GMT NIS2 in practice for Italian SMEs and MSPs: the 2026 ACN compliance calendar, the categorisation window closing at the end of June, incident notification and baseline measures. What to do now and where CyberScan supports gap analysis. Cybersecurity Compliance Governance https://www.noze.it/en/insights/fhir-llm-digital-health-on-premise/ https://www.noze.it/en/insights/fhir-llm-digital-health-on-premise/ Mon, 25 May 2026 00:00:00 GMT On-premise clinical RAG built on FHIR as the interoperability layer, Open Source biomedical LLMs and data governance. EHDS roadmap towards 2027 and the AI Act for medical devices, with a link to AgenticHealth. Digital Health AI Open Source Compliance Governance https://www.noze.it/en/insights/open-intelligence-secure-governance-agentic/ https://www.noze.it/en/insights/open-intelligence-secure-governance-agentic/ Fri, 22 May 2026 00:00:00 GMT In the era of autonomous agents and multi-agent orchestration, governance is not a layer to bolt on afterwards but a by-design property. The Open Source opening of Admina is concrete proof of the OISG thesis. AI Governance Open Source Cybersecurity Compliance noze https://www.noze.it/en/insights/admina-open-source/ https://www.noze.it/en/insights/admina-open-source/ Wed, 20 May 2026 00:00:00 GMT Admina v0.9.1 released (20 May 2026): Apache 2.0 Open Source framework for AI governance, created by Stefano Noferi and sponsored by noze. Python SDK + Rust proxy, 4 governance domains, ~6µs overhead, 700+ tests. AI Open Source Compliance noze Governance https://www.noze.it/en/insights/ai-act-digital-omnibus-2026/ https://www.noze.it/en/insights/ai-act-digital-omnibus-2026/ Fri, 08 May 2026 00:00:00 GMT On 7 May 2026 the European Parliament and Council reached a provisional political agreement on the Digital Omnibus on AI: high-risk stand-alone systems pushed to 2 December 2027, exemptions extended to small mid-caps (up to 500 employees), new ban on non-consensual sexual deepfakes and CSAM. AI Compliance Governance https://www.noze.it/en/insights/copy-fail-cve-2026-31431/ https://www.noze.it/en/insights/copy-fail-cve-2026-31431/ Thu, 30 Apr 2026 00:00:00 GMT CVE-2026-31431 'Copy Fail': nine years of Linux kernel (2017–2026) exposed to a 4-byte out-of-bounds write at the intersection of AF_ALG, splice() and the authencesn algorithm. A 732-byte Python payload is enough to tamper with /usr/bin/su and gain root. Upstream patch already shipped. Open Source Cybersecurity https://www.noze.it/en/insights/llm-wiki/ https://www.noze.it/en/insights/llm-wiki/ Tue, 28 Apr 2026 00:00:00 GMT The 'LLM Wiki' pattern (proposed by Andrej Karpathy in a gist on 4 April 2026) replaces ad-hoc RAG regeneration with a persistent Markdown wiki incrementally maintained by an LLM. Three layers (sources, wiki, schema), ingest/query/lint workflow, tooling: Obsidian, Git, qmd. Open Source AI https://www.noze.it/en/insights/deepseek-v4-preview/ https://www.noze.it/en/insights/deepseek-v4-preview/ Fri, 24 Apr 2026 00:00:00 GMT DeepSeek ships V4 Preview with a 1M-token context window, two MoE variants (V4-Pro 1.6T total / 49B active and V4-Flash 284B / 13B active), a new sparse attention design and API compatibility with OpenAI and Anthropic. Open Source AI https://www.noze.it/en/insights/ubuntu-26-04-lts/ https://www.noze.it/en/insights/ubuntu-26-04-lts/ Thu, 23 Apr 2026 00:00:00 GMT Canonical releases Ubuntu 26.04 LTS 'Resolute Raccoon': Linux kernel 7.0, GNOME on Wayland, Rust user-space utilities, confidential computing, integrated AI stack and Livepatch on Arm64. Open Source https://www.noze.it/en/insights/qwen-3-6/ https://www.noze.it/en/insights/qwen-3-6/ Wed, 22 Apr 2026 00:00:00 GMT Alibaba opens the Qwen3.6 cycle with Plus (hosted, April 2), 35B-A3B (open on Hugging Face and ModelScope, April 17) and Max-Preview. Stated focus on agentic coding, multimodal reasoning and a 1M-token context. Open Source AI https://www.noze.it/en/insights/oisg-paradigm-autonomous-ai/ https://www.noze.it/en/insights/oisg-paradigm-autonomous-ai/ Thu, 16 Apr 2026 00:00:00 GMT noze adopts OISG (Open, Intelligent, Secure, Governed), the architectural paradigm proposed by Stefano Noferi for the design and governance of autonomous AI systems. Four interdependent pillars that translate into concrete benefits for products, solutions, partners and clients. AI Cybersecurity Compliance Open Source noze https://www.noze.it/en/insights/innovation-management-r-and-d-industry-4-5/ https://www.noze.it/en/insights/innovation-management-r-and-d-industry-4-5/ Mon, 13 Apr 2026 00:00:00 GMT How innovation management, research and development and the Industry 4.0 and 5.0 paradigms intersect to guide the technological transformation of organisations. R&D https://www.noze.it/en/insights/linux-kernel-7/ https://www.noze.it/en/insights/linux-kernel-7/ Sun, 12 Apr 2026 00:00:00 GMT The Linux kernel moves to version 7.0. No technical revolution: it is a numbering choice. But under the surface, Rust, io_uring and support for emerging architectures are consolidating. Open Source https://www.noze.it/en/insights/ciso-automated-vulnerability-assessment/ https://www.noze.it/en/insights/ciso-automated-vulnerability-assessment/ Wed, 08 Apr 2026 00:00:00 GMT IBM, Verizon and Ponemon data prove that security automation slashes breach time and costs. Here is how a CISO can act now. Cybersecurity Compliance https://www.noze.it/en/insights/dpo-automated-compliance-gdpr-ai-act/ https://www.noze.it/en/insights/dpo-automated-compliance-gdpr-ai-act/ Tue, 07 Apr 2026 00:00:00 GMT With over EUR 4.5 billion in GDPR fines issued and the AI Act in force, DPOs need integrated tools. The numbers speak for themselves. Legal Compliance Governance https://www.noze.it/en/insights/compliance-manager-integrated-platform/ https://www.noze.it/en/insights/compliance-manager-integrated-platform/ Mon, 06 Apr 2026 00:00:00 GMT The Ponemon Institute certifies it: investing in an integrated compliance platform costs 2.7 times less than bearing the consequences. Numbers, trends and solutions. Compliance AI Governance https://www.noze.it/en/insights/cyberscan-launch/ https://www.noze.it/en/insights/cyberscan-launch/ Wed, 01 Apr 2026 00:00:00 GMT noze announces the launch of CyberScan, a cloud vulnerability assessment and NIS2 compliance platform for Italian SMEs. Access by invitation and on request, then verified registration. Cybersecurity Compliance noze https://www.noze.it/en/insights/ehds-secondary-use-2027/ https://www.noze.it/en/insights/ehds-secondary-use-2027/ Wed, 18 Mar 2026 00:00:00 GMT One year after the EHDS Regulation entered into force: the state of preparedness for the first obligations (March 2027), the construction of national Health Data Access Bodies, the work on data permit procedures and on HealthData@EU infrastructure. Legal Digital Health Compliance https://www.noze.it/en/insights/eu-ai-act-what-changes/ https://www.noze.it/en/insights/eu-ai-act-what-changes/ Sun, 15 Mar 2026 00:00:00 GMT A clear overview of the new EU rules on artificial intelligence and what companies need to do before the deadline. Legal AI Compliance Governance https://www.noze.it/en/insights/opencode-remote-ops-ssh/ https://www.noze.it/en/insights/opencode-remote-ops-ssh/ Thu, 05 Mar 2026 00:00:00 GMT Using opencode (opencode.ai) inside an SSH session for log triage, package audit and small maintenance tasks on Linux servers. Human-approval flow, no secrets in prompts, preventive backup. Open Source AI Tutorial https://www.noze.it/en/insights/medical-software-ai-certifications-global/ https://www.noze.it/en/insights/medical-software-ai-certifications-global/ Mon, 16 Feb 2026 00:00:00 GMT Overview of software as a medical device certifications worldwide, spanning EU MDR, FDA, MHRA and the regulatory challenge posed by the growing adoption of artificial intelligence. Legal Digital Health Compliance AI R&D https://www.noze.it/en/insights/gb10-ai-workstation-on-premise/ https://www.noze.it/en/insights/gb10-ai-workstation-on-premise/ Tue, 20 Jan 2026 00:00:00 GMT NVIDIA GB10 Grace Blackwell: desktop workstations (DGX Spark, ThinkStation PGX) with 128 GB unified memory and 1 PFLOP FP4 for up-to-200B models on-premise. AI R&D noze Cybersecurity Digital Health Hardware https://www.noze.it/en/insights/open-intelligence-secure-governance/ https://www.noze.it/en/insights/open-intelligence-secure-governance/ Sun, 14 Dec 2025 00:00:00 GMT The payoff that gives a name to over 25 years of AI and Cybersecurity built together, by the same team, on the same projects. noze AI Cybersecurity Digital Health Web Open Source R&D Governance https://www.noze.it/en/insights/openehr-fhir-cda-comparison/ https://www.noze.it/en/insights/openehr-fhir-cda-comparison/ Wed, 22 Oct 2025 00:00:00 GMT Three different models for representing healthcare data: CDA R2 for the archival document, FHIR for transactional exchange, openEHR for native persistence with clinical governance. Comparison on data model, query, governance and adoption. Digital Health R&D https://www.noze.it/en/insights/mlflow-serving/ https://www.noze.it/en/insights/mlflow-serving/ Mon, 15 Sep 2025 00:00:00 GMT MLflow 2.x consolidates the ML lifecycle: tracking, model registry, deployment, evaluation. MLflow Deployments for REST serving, Kubernetes and cloud provider integrations. Open Source AI https://www.noze.it/en/insights/clinical-rag-open-source-llm/ https://www.noze.it/en/insights/clinical-rag-open-source-llm/ Tue, 08 Jul 2025 00:00:00 GMT Retrieval-Augmented Generation applied to diagnostic support with Open Source LLMs (Llama, Mistral, Gemma, DeepSeek, BioMistral, Meditron), integration with FHIR and DICOM, on-premise deployment, EU AI Act and MDR compliance. Digital Health R&D Open Source AI https://www.noze.it/en/insights/git-20-anni/ https://www.noze.it/en/insights/git-20-anni/ Sun, 15 Jun 2025 00:00:00 GMT Twenty years since Linus Torvalds' first commit (April 2005). From kernel development tool to de facto standard: GitHub, GitLab, worktree, sparse checkout and partial clone. Open Source https://www.noze.it/en/insights/ehds-regulation-2025-327/ https://www.noze.it/en/insights/ehds-regulation-2025-327/ Wed, 16 Apr 2025 00:00:00 GMT The European Health Data Space Regulation, published on 5 March 2025 and in force from 26 March 2025. Primary use, secondary use, Health Data Access Bodies, MyHealth@EU, HealthData@EU and the application calendar. Legal Digital Health Compliance https://www.noze.it/en/insights/apache-httpd-30-anni/ https://www.noze.it/en/insights/apache-httpd-30-anni/ Tue, 15 Apr 2025 00:00:00 GMT Thirty years since the first release of Apache HTTP Server (1995). From patches on NCSA httpd to founding project of the Apache Software Foundation: HTTP/2, event MPM, mod_proxy and the Nginx challenge. Open Source Web https://www.noze.it/en/insights/a2a-agent-to-agent/ https://www.noze.it/en/insights/a2a-agent-to-agent/ Wed, 09 Apr 2025 00:00:00 GMT Google Cloud announced A2A at Next '25 on 9 April 2025: an open Apache 2.0 protocol for interoperability between heterogeneous agents. Agent Card, task lifecycle, HTTP + SSE messaging, 50+ partners. Open Source AI https://www.noze.it/en/insights/goose-mcp-sanita-fhir/ https://www.noze.it/en/insights/goose-mcp-sanita-fhir/ Tue, 18 Mar 2025 00:00:00 GMT Block's Goose configured with a custom MCP server exposing read-only metadata of an anonymised FHIR bundle. Local LLM via Ollama, directory isolation, use of test datasets. Open Source AI Tutorial