The German BMI has published a 691-page Referentenentwurf (5 July 2026, currently in interministerial coordination, with stakeholder comments requested) that rewrites German intelligence law from the ground up: a new BVerfSchG, a new BNDG, and a standalone statute for the Unabhängiger Kontrollrat (UKRat) for the first time, plus sixteen amendments to laws from the Vereinsgesetz to the Abgabenordnung.

It makes some very strange moves. I’ll go through a few here.

Whoever drafts this is drafting a toolkit on the assumption the AfD will never hold the Innenministerium. They are standing up a domestic deception-and-intervention license, with oversight consolidated into a single body whose pre-approval the agency head can defer on self-certified urgency. I said defer, yes, technically not a bypass, because the order lapses if unconfirmed. But come on people, in a covert fire-and-forget the whole oversight bureaucracy becomes meaningless: false information already injected and data deleted are galloping horses after a barn door was left open. The AfD are probably already strapping on their spurs in anticipation.

Germany currently has a Kremlin-aligned Nazi-adjacent party at or near the top of federal polls. The Weimar lesson for anyone paying attention, precisely stated, is not that the state was too weak or too strong, it’s that defenders of the constitution built instruments that were handed intact to its enemies. The Gestapo didn’t build the Prussian political police; it inherited it via Preußenschlag.

Vor allem die Polizeipräsidenten werden ausgetauscht. Hitlers SA- und SS-Mannen haben keinen Grund mehr, die preußische Polizei zu fürchten.

EN: (It is primarily the police chiefs who are being replaced. Hitler’s SA and SS men no longer have any reason to fear the Prussian police.)

The engineering failures of this document are what make the risks of inheritance catastrophic. A well-instrumented system constrains anyone and everyone using it. We always used to call it hard to use wrong, easy to use right. The BMI document is fail-open, like a loaded weapon handed out with the safety filed off and a rainbow colored “do no evil” sticker on it.

Look at the new “efficiency” claims, for example. On paper it looks like there is oversight because the draft extends pre-approval to more measure types and even keeps the BfDI’s name on a section header. However, oversight means redundancy, and there’s none of that. Three independent overseers (G10-Kommission, BfDI, UKRat) compress into a single one. The G10-Kommission, created in 1968 as the constitutional price of restricting Art. 10, is simply gone, its elimination booked as budget savings. Redundant, overlapping overseers are the spine of oversight, meaning the opposite of a waste: they are upstream investments that cross-check each other and can’t all be captured or starved at once, which prevents downstream costly disasters. When your internal and external auditor are the same person reporting to themselves, you are talking Enron (I managed a five-state Computer Risk practice at Arthur Andersen, I can tell you all about it). The draft has a single body funded at just €8.86M a year to control an apparatus spending €269M a year on BfV IT operations alone. Thirty to one, capability to control, as if to say control can be devalued when it’s the valuation lever on the capability. Penny wise, pound foolish.

Look at the history of Germany, for another example. The Trennungsgebot exists because of the Polizeibrief and the Gestapo precedent, where an agency both watched and acted, covertly, without judicial process. This document has zero mentions of Gestapo, Stasi, or the historical rationale for keeping separation. The draft cites only post-2013 BVerfG doctrine, as if the principle were a data-protection technicality rather than a lesson in history called “never forget”.

Perhaps the most peculiar move of all, even more than Germany refusing to acknowledge Nazi precedent, is that automated hackback is justified with “human review adds no relevant quality assurance.”

Ein Zwischenschritt menschlicher Bearbeitung leistet hier keine relevante Qualitätssicherung, verzögert aber Abwehr erfolgsgefährdend.

EN: (An intermediate step involving human processing provides no relevant quality assurance here, yet delays the response to a degree that jeopardizes success.)

That’s backwards, per the canon of hackback and per automation safety. Horseshit, in a word. Adversaries who map an automated response own it: learn the trigger conditions and you can redirect state countermeasures into friendly fire. The draft gestures at quality (accuracy, robustness, cybersecurity, criteria lifted from the EU AI Act while exempting itself from the EU AI Act) and has no quantification. No error tolerance, no attribution-confidence threshold, no third-party-harm analysis, nothing. The automated measures themselves, traffic redirection and data deletion, are placed in a shared-infrastructure category where misattribution costs hit innocent bystanders: “attacker” data wiped on a compromised host, and a victim’s server just got destroyed. These are very, very old talking points in hackback that get zero attention: we litigated this on my blog in 2013 and yet the draft adopts the losing side without the debate. The scope limitation to cyberattack contexts is in fact the shared infrastructure, contested attribution, and machine-speed response. That’s automation in the wild west where you want it the least.

Notably, the word Desinformation is used only to describe adversaries (e.g. Russia), never German measures. When the domestic intelligence service does it (§ 60(2)(1)(c), feeding false information through informants into networks to steer behavior) it gets branded “Schutzmaßnahme” instead. But it’s the same thing. This reminds me how way back before there was a Bundesrepublik, Nicolai’s Abteilung IIIb ran Desinformation as Aufklärung. When the euphemism appears here I see a domestic deception charter, license to perform disinformation, for the agency whose remit is observing political movements.

Germany just wrote state disinformation into statute. Germany. Of all places. Germany.

Back in 2012 I openly led hackback around the world, and I even gave a 2023 lecture titled “The Heaviest of Burdens: Hackback” at the National Security Seminar, William and Mary Law School. Somewhere around here I have the gun club t-shirt to prove it. I’m not the typical civil liberties voice here, in other words. As a long-time leading hackback advocate, although arguably a salty dog, this document smells like malpractice.

Fight the anti-constitutional party through the public, judicial instrument built for exactly this, Article 21, precisely so Germany never again needs a secret political police. Right? Is this thing on? In 1932 the judicial remedy failed because it arrived after the apparatus changed hands. The transparency of well-engineered fail-safe instruments is what survives hostile takeovers, while the rushed self-certified covert stuff like this always gets studied later as a blunt takeover instrument.