Blog
TOPIC Two Models in One Day On September 22, 2026, Anthropic and OpenAI released new models within one to two hours of each other. The real story is not the comparison between them, but a measurement published the same day: top-end performance is collapsing in price.
Read →
TOPIC Beck Becomes Majority Owner of Noxtua On 23 September 2026 it emerged that C.H.Beck is becoming majority shareholder of legal AI provider Noxtua, in a round of more than 100 million euros. What the new market monitor actually reports, and the four questions a vendor has to answer.
Read →
TOPIC Software Is a Product Directive (EU) 2024/2853 turns software and AI into products for liability purposes — strict liability, whether local, cloud or SaaS. What that means for contracts and insurance.
Read →
TOPIC When Enforcing Rights Gets Cheap By mid-August 2026 the Arbeitsgericht Berlin had recorded some 16,000 cases — as many as in all of 2024. What the numbers prove, what they do not, and how companies should respond to waves of near-identical AI-drafted claims.
Read →
TOPIC Buying Sovereignty Three billion euros for a French provider, EU data residency as a bookable region, a fully open Swiss model — and EU data centres that will not run before 2028. What holds up today.
Read →
TOPIC Efficiency versus Explainability OpenAI's new flagship reportedly buys its leap with an architecture that no longer records the path to a result in readable language. Why that is a selection criterion for anyone who has to justify decisions.
Read →
TOPIC GPT-6 Astra and the AGI Question OpenAI has unveiled GPT-6 Astra and spoken of an AGI era. The sharpest objection sits in the evaluator's own record post: 99.9 percent was reached only in a setup where the model was allowed to write its own solver scripts — in the standard setup it is 62.7 percent.
Read →
TOPIC AI Labelling Becomes a Product Feature Since 1 September 2026, the outputs of certain models carry a text watermark and files from code execution carry C2PA Content Credentials. What Article 50 of the AI Act actually requires, whom it binds, and why 2 December 2026 is the most inconspicuous important date of the year.
Read →
TOPIC 88 Percent Use AI, 6 Percent Earn… McKinsey counts 88 percent AI users, the ifo Institute 54.5, the KfW just 20. Why reputable surveys diverge this far — and which number is the only one that matters for a mid-sized company.
Read →
TOPIC Agent Skills in Practice Agent Skills left beta on August 19, 2026. What a stored work instruction actually delivers, which use cases pay off in a mid-sized company — and when the effort outweighs the benefit.
Read →
TOPIC From Answering to Acting Since August 19, 2026, computer use and a dedicated browser tool have been generally available. The wrong answer becomes the wrong action — and a judgment by the OLG Hamm shows where attribution is heading.
Read →
TOPIC The MCP Protocol Broke on July 28 The MCP specification of July 28, 2026 makes the protocol stateless, declares server/discover mandatory and deprecates HTTP+SSE. Anyone who integrated in 2025 has to touch it — and needs a budget line most projects do not have.
Read →
TOPIC The Attack Did Not Come Through the… On August 4, 2026 an npm worm distributed malicious code through a package with 127 million weekly downloads — and specifically went after the configuration files of AI coding tools. Why the credentials do not sit in the model but in unremarkable files on developer machines.
Read →
TOPIC GEMA v. Suno On 31 July 2026 the Landgericht München I held that training on protected works is an act of reproduction and that the text-and-data-mining exception does not cover it. The judgment is not final — the two questions it raises face companies now.
Read →
TOPIC The Bundesnetzagentur Is Now Your AI… Since July 29, 2026, Germany has had a statutory AI supervisory authority for the first time: the KI-MIG. Who is now competent, where you report and ask — and why the widely quoted EUR 50,000 under Section 15 KI-MIG is not a fine ceiling.
Read →
TOPIC The AI Act, Cut in Half 2 August 2026 arrived — but only halfway. An amending regulation has postponed the high-risk regime to December 2027 and softened the AI literacy duty. What applies now, what is still waiting, and why the EU openly admits its own delay.
Read →
TOPIC Claude Opus 5 Anthropic has released Claude Opus 5 — its most capable model yet, at the same price as its predecessor. A neutral overview of the key facts, pricing and the benchmark picture: this time broadly confirmed by independent measurement — and why, of all things, reliability points the other way.
Read →
TOPIC GPT-5.6 OpenAI has released GPT-5.6 in three variants: Sol, Terra and Luna. A neutral overview of the new naming scheme, availability and pricing, the benchmark picture — and why the independent safety test casts doubt on the record numbers.
Read →
TOPIC Context Lock-In in AI It's not the model that ties you to an AI provider — it's the context around it: prompts, memory, agents, RAG. Why most migrations fail, what that has to do with Anthropic, and how four steps protect you.
Read →
TOPIC Fable 5 Is Back Anthropic redeployed Claude Fable 5 worldwide on 1 July 2026 — secured by a new classifier. What it actually does, where it stays leaky, and why it matters for your own control layer.
Read →
TOPIC The Claude Myth Fact versus marketing: what Constitutional AI actually is, what companies can realistically expect from Claude — and why lawyers in particular should look closely.
Read →
TOPIC Fable 5 & Mythos Shut Down Anthropic shut down Claude Fable 5 worldwide after a US government directive. What that reveals about vendor lock-in, liability, and migration strategy.
Read →
TOPIC AI Agents in the Enterprise AI agents act autonomously: real value, new liability questions. A guide from the business lawyer who builds the agents - use cases, governance, liability.
Read →
TOPIC AI Contracts A solid AI contract governs five points: performance, liability for hallucinations, IP in the output, data protection/DPA, and the SLA. What to negotiate in — from a commercial lawyer who also builds the AI solution. As of: June 2026.
Read →
TOPIC Using AI in Customer Service… Using AI in customer service compliantly: transparency obligation (Art. 50 EU AI Act since 08/2026), liability (OLG Hamm) & GDPR — a duties checklist from a business lawyer.
Read →
TOPIC AI for Law Firms Using AI in law firms safely: what sec. 203 StGB & the BRAO permit, what the BRAK guidance requires, liability for AI hallucinations — by a business lawyer.
Read →
TOPIC DPIA for AI Systems When is a DPIA mandatory for AI — and when is it not? Threshold analysis, a 6-step process and DPIA vs. FRIA, explained in practical terms by a business lawyer.
Read →
TOPIC Safeguard AI Hallucinations AI hallucinations are a liability risk — not just a bug. Here is how to design a human control layer that satisfies Art. 14 of the EU AI Act and Art. 22 GDPR.
Read →
TOPIC EU AI Act & GDPR The AI Act's concrete obligations, deadlines and risk classes, its interface with the GDPR, and a tight action checklist — from the combined perspective of lawyer and technologist.
Read →
TOPIC Implementing the AI Literacy… Art. 4 of the EU AI Act calls for measures supporting AI literacy. How to implement it: content, roles, a training framework, documentation — by a business lawyer.
Read →
TOPIC Process Automation with AI Where to actually begin with AI process automation, how to calculate ROI properly, and the legal question up front that decides between success and an expensive failure.
Read →
TOPIC LLM Wrappers Is your GPT wrapper (AI wrapper) a real product or just a thin prompt layer? Defensibility, data layer, moat — plus the legal layer (liability, GDPR/DPA, AI Act, IP) that tech articles leave out. As of April 2026.
Read →
TOPIC AI Use Cases for SMEs 10 AI use cases for SMEs with effort, benefit, time to ROI, and a solid GDPR / EU AI Act assessment — written by a business lawyer who builds the solutions himself.
Read →
TOPIC Open-Source vs. Proprietary LLMs Open-source LLM or proprietary API? A decision guide with an honest TCO/break-even calculation, licensing traps (Llama, Mistral) and a GDPR assessment.
Read →
TOPIC How to Set Up an AI Pilot Project the… Setting up an AI pilot project the right way means testing a clear hypothesis within a few weeks using real data and hard success criteria — before you invest. A six-step guide, including GDPR and the EU AI Act.
Read →
TOPIC Evaluating AI Outputs AI quality can't be guessed. Here's how to make LLM outputs measurable with evals — golden datasets, metrics, LLM-as-a-judge — and use them as compliance proof.
Read →
TOPIC Privacy in the Prompt Where does a prompt travel, is it used for training, do you need a data processing agreement — and why shadow AI is the real risk. Tech and GDPR explained in one piece.
Read →
TOPIC Avoiding AI Vendor Lock-In How to become independent of individual AI vendors: model-agnostic architecture, watertight exit clauses, and your switching rights under the EU Data Act — explained from both a technical and a legal perspective, from a single source.
Read →
TOPIC Understanding Context Windows & Token… What do 1 million tokens mean for cost, practice, and data protection? A clear guide with conversion and cost tables — technically sound and legally grounded.
Read →
TOPIC MVP for Startups MVP development for startups: cost, timeline, no-code vs. custom — and how to build fast without piling up technical AND legal debt. From a lawyer who codes.
Read →
TOPIC RAG vs. Fine-Tuning for Enterprises RAG or fine-tuning? The decision guide for enterprises — with a comparison table, cost ranges, maintenance effort, and the GDPR perspective (German DSK guidance on RAG, BfDI). From a business lawyer who builds AI solutions himself.
Read →
TOPIC Build vs. Buy Build vs. Buy: custom software or SaaS? The decision framework with four axes — including TCO, vendor lock-in & GDPR. From a lawyer who builds software himself.
Read →
TOPIC From Prompt Engineering to Context… Context engineering is supplanting prompt engineering as the central discipline. Definition, the key difference, the new lever (RAG, tools, memory, context window) — plus the overlooked GDPR dimension for businesses.
Read →
TOPIC Business Lawyer + Developer in One… AI projects fail at the seam between law and IT. Why combining business lawyer and developer in one person closes that seam — the concrete value, explained honestly.
Read →
TOPIC EU-Hosted vs. US LLMs US model, EU provider, or on-premise? The substantive comparison on data sovereignty, the CLOUD Act vs. Art. 48 GDPR, and local LLMs — from a business lawyer who also builds the solution. With a decision framework and cost/effort comparison. As of December 2025.
Read →
TOPIC Shadow AI in the Workplace Shadow AI: employees use ChatGPT & co. without approval. Why it touches GDPR, trade secrets and the AI Act – and how to rein it in organizationally and technically. With an action plan, data and sources.
Read →
TOPIC AI Policy for Your Company What a corporate AI policy must accomplish, what belongs in it (with a building-block overview), and how to roll it out effectively — legally precise and technically enforceable. A guide, not case-specific legal advice.
Read →