Last Saturday this site put three letters on the wall: (a) permanent voluntary scanning at full scope with no review, (b) permanent voluntary scanning with a review layer, (c) private messages out of scope conditionally. Tuesday's sixth trilogue on the Child Sexual Abuse Regulation ended after midnight with none of them. No political agreement on detection. That is the first thing to report, and the honest scoring of a prediction post includes the branch nobody drafted: no deal at all.

What follows is what actually happened, from the sources that can be checked — Patrick Breyer's trilogue report, published the morning after, and the Parliament's press release on the EU Centre's tasks.

Scorecard: the letters

(a) — blocked, for now. The Irish presidency came with what Breyer's office calls "Search Plans": providers may choose to scan non-public content; the scanning can cover large parts of a service or entire chat groups rather than individual suspects; a submitted search plan takes effect automatically unless an authority actively vetoes it — silent approval; the whole apparatus would be made permanent where Regulation 2026/1881 sunsets in April 2028; and provider-initiated scanning would be smuggled into enforceable risk-mitigation rules, meaning platforms scan to avoid fines. Parliament's negotiating team refused the package, holding that scanning of private communications must be limited to persons actually linked to child sexual abuse. That is the entire difference between (a) and Parliament's mandate, and it held.

(b) — deferred, not resolved. The search-plan-with-review construct from the September 18 technical meeting moved to October's expert talks. Nothing was agreed. It remains the only mechanism ever tabled that puts a national judicial or independent administrative authority in front of what a provider scans.

(c) — unanswered. The presidency's question to the 27 ambassadors — whether to exclude private messages from the permanent regulation if no compromise matches "the current voluntary practice" — produced no published answer. The question survives into the next round. Remember what it is attached to: an effectiveness floor defined as report volume, which nobody can measure publicly until providers owe their first error-rate reports on February 1, 2027. The floor is still being set before the first public numbers exist.

One honest note on the pre-registration: last Saturday framed Tuesday as deciding which letter survives. The correct reading now is that none did, and the reason none did is mechanical — the Council would not move off voluntary mass scanning, Parliament would not accept it, and the presidency judged a partial deal worse than a public stalemate. Buying a few weeks is a real outcome. It is not a settlement.

The technical win: mitigation measures

The substantive concession: the text will be clarified so that authorities cannot enforce voluntary scanning of private communications through the back door of mitigation measures. This was the sharpest tooth in the Council's September paper — the one that turned "voluntary" into something a platform could be fined into doing. Closing it in writing matters for operators reading procurement clauses: the difference between a provider that may scan by choice and one that a regulator can compel by penalty is a difference you can, in principle, test when you ask a hosting vendor what their scanning posture is. Weak, but real, and it was under the door until Tuesday's session.

What actually got agreed: the public-content machinery

The drafted agreement arrived on the other basket, and it is the part this site's readers should read twice.

Negotiators reached a tentative deal on detecting CSAM in publicly accessible content:

  • Detection orders go to hosting service providers — including social media platforms and cloud providers — and can be issued by administrative authorities, not only courts.
  • The obligation is to continuously monitor all public uploads, not just uploads from users suspected of a crime. Parliament's own earlier proposal had limited public orders to suspected users. That limit did not survive.
  • Orders can be issued to any hosting provider regardless of size — there is no small-provider carve-out in the drafted text.
  • Separately, the EU Centre gets its task list: own-initiative crawling of public spaces for known and new CSAM. Both baskets of content. The Parliament press release frames this as the rapporteur's achievement; it is also the general-purpose scanning infrastructure built and funded as law, whatever happens to private messages in November.

Read that against the operational analysis from September 20: the constant there was that infrastructure outlives mandates. The machinery is now further along than the mandate — drafted, with orders, authorities, and the Centre's public-content crawl provisionally agreed, while the private-message basket remains in dispute.

There is an irony worth naming without pretending it settles anything. The DSA's Article 8(1) prohibits imposing general monitoring obligations — generic scans of everything without individual assessment. Parliament's team held the line against suspicionless scanning of private messages, then drafted an agreement requiring continuous monitoring of all public uploads regardless of suspicion. That is the shape of a general monitoring obligation for public content, issued by an administrative authority. The line against generic scanning did not get repealed; it got moved to whichever edge negotiators could move it. If your community's public content lives on someone else's cloud, the moved edge is where you live.

What this means for community operators

Concretely, for the fire hall, the city office, the co-op, the collective:

  • The public/private split is becoming load-bearing law. Everything marked public on your stack — public Matrix rooms (world-readable shared history, federated to anyone who joins), public Nextcloud shares, public web hosting — sits in the drafted basket. Private spaces — encrypted DMs, private rooms, personal files — sit in the contested one. The drafting of public content is not contingent on November; it survived Tuesday and will likely survive whatever November produces. Plan your data layout accordingly: public content on your own disk faces orders you can answer for; public content on a rented platform faces the platform's compliance with authorities you do not talk to.
  • Size is not a shield. "Regardless of size" is in the text. The common assumption that community-scale operators are beneath the threshold of these obligations has now been tested twice — once with NIS2 and once here — and lost twice. Small is a budget fact, not a compliance fact.
  • The private-message fight carries into November with the same trap. The Council's standing fallback is to keep Chat Control 1.0 — the voluntary mass-scanning regime the Council re-enacted in July with one vote against and one abstention — as a loophole if Parliament keeps refusing detection orders on private content. The evidence that the regime is a failure is not ideological: German police data shows over half of investigations triggered by scanning target minors themselves for consensual sexting, and roughly 75% of flagged chats are not actionable. A floor defined as preserving that regime's "effectiveness" is preserving its report volume, not its results.
  • The dates still decide the fight. Technical talks on non-public detection run through October. The trilogue expected in November is the final attempt at a political deal. February 1, 2027 is when the first public error-rate numbers become available — after the floor is set, before the regulation is final, in time to embarrass no one who has already voted. April 3, 2028 is when 1.0 sunsets unless the November deal keeps it alive as a parallel track.

The architecture answer, unchanged

Saturday's post closed: run the homeserver, own the disk, the only fixed point is the one you hold. The trilogue confirmed the constant rather than moving it. If November lands a compromise that keeps private messages in scope in any of the Council's three variants — permanent voluntary, review-layer voluntary, or detection orders without courts — the operator argument gets steeper but does not change: messages that never transit a provider's scanning surface are covered by no order and no search plan. If (c) eventually wins, the permanent law stops short of private messages, scanning continues under 2026/1881 anyway until 2028, and each future extension has to win a vote again — which is the whole point of preferring a lapsing regime to a permanent one.

What November decides is who holds the scanning surface for private messages. What Tuesday already decided is who holds it for public ones. Either way, the answer to both was always the same key: your own.