Last week this site argued that the scanning decision has moved to whoever operates the server — because the Council's own lawyers read the voluntary model as leaving scanning inside end-to-end encryption untouched. That argument now has a new piece of evidence attached, and a new question to track.

The question is literally on the table.

The document

Document 13158/26, dated 18 September 2026, "Preparation for the trilogue." LIMITE, as everything on this file is — the Council's public register lists it by title only. The former MEP Patrick Breyer published the full five pages; Agence Europe reported on it on September 22. It prepared the ambassadors' meeting of September 23 and ends with three questions put to the 27 governments "in preparation of the trilogue on 29 September 2026":

a) detection orders on publicly accessible content for known CSAM, plus own-initiative searches by providers on non-public content — private messages — for known and new CSAM and the solicitation of children. Read plainly: write the providers' voluntary scanning of private messages into permanent law, at the interim regulation's full scope.

b) the compromise: own-initiative searches on non-public content would require a search plan reviewed by a national judicial or independent administrative authority, and providers would be obliged to use an indicator database provided by the EU Centre.

c) the fallback: "If no compromise solution can be agreed providing at least for the same level of effectiveness as the current voluntary practice, would you accept to exclude own-initiative searches or detection orders on non-public content from the scope of this Regulation?"

Question (c) is new in kind, not just degree. For years the Council's opening position was mandatory detection orders; since November 2025 it has been voluntary scanning made permanent. Now the presidency has formally asked the governments whether private messages should be outside the permanent regulation altogether. Whatever the answer, the fact that the question had to be asked describes the negotiation: five trilogue rounds have not settled detection, and the presidency is preparing its delegations for the possibility that no compromise holds.

What (c) would change — and what it would not

If (c) landed, private messages would sit outside the permanent regulation: no detection orders against them, and no permanent legal basis in this regulation for providers to scan them. Public content stays in — detection orders there are limited to known CSAM, explicitly to avoid a general monitoring obligation. The EU Centre gets built anyway. The note records technical agreement on the Centre's own-initiative searches of publicly accessible content, and says the presidency intends to launch the process for choosing the Centre's seat if the trilogue confirms it.

It would not end the scanning of private messages. Regulation (EU) 2026/1881 — Chat Control 1.0 — legalizes voluntary scanning until 3 April 2028. After that, the derogation needs extending again (it has been twice) or a new legal basis. And (c) is a question, not a decision: Coreper publishes no outcome for an item like this, and the governments' answers are not public. What the ambassadors settled on September 23 shows up only in the position the presidency carries into the trilogue on Tuesday.

The condition is the story

Read the condition attached to (c) again: "providing at least for the same level of effectiveness as the current voluntary practice."

The presidency's earlier state-of-play note, document 11956/26 of September 10, states the bottom line plainly: "the Council does not want a permanent framework that achieves less than the voluntary regime." Effectiveness, in Council usage, means report volume — the presidency has consistently communicated the need to "maintain the current level of reports."

Now line that up with the calendar the interim regulation itself sets:

  • October 1, 2026 — the Commission must publish the list of organisations that receive reports about Europeans.
  • February 1, 2027 — providers owe their first public reports on error rates and false positives.
  • April 1, 2027 — the grace period ends for providers already scanning before July 31, 2026 without a completed data protection consultation.

The Council wants to preserve "current effectiveness" before the first public numbers on what current scanning actually produces exist. Nobody outside the providers knows the error rates yet. The first honest measurement of the voluntary regime is due February 1 — months after the trilogue, and likely after Coreper endorses whatever text survives it. A floor built on an unpublished metric is not a floor. It is a negotiation position wearing one.

What the state of play shows about the room

The same note records the replies to the July reflection paper: 14 member states responded, and "most Member States responding are in principle opposed to the concept of targeting detection to specific users." The presidency's own text goes further: requiring searches to target specific users "would not generate any added value compared to the possibilities already existing under criminal law."

That is the presidency dismissing, in writing, the only detection model the Parliament's mandate permits — judge-issued orders against named suspects. Parliament has already shown what it does with that: it refused to extend the temporary regime in March, and on July 9 a rejection motion drew 314 votes, short of the 360 threshold only because abstentions and empty seats count against it. At the September 18 technical meeting Parliament tested a different instrument — search plans describing the technologies used and the safeguards, reviewed by a judicial or independent administrative authority — which is exactly question (b).

So the live options going into Tuesday are: (a) permanent voluntary scanning at full scope with no review; (b) permanent voluntary scanning with a review layer and the EU Centre's database; (c) private messages out of scope, conditionally. The sequencing in the note matters too: detection orders on private content can be discussed "only once there is agreement on the parameters and the scope of the own-initiative searches." Searches first, orders later — if ever.

What an operator does with this

Three scenarios, one constant.

If (a) wins, the scanning decision for hosted platforms is written into permanent law with no oversight at all. The operator argument gets harder to make at the margin, because "everyone does it" acquires a legislative footnote.

If (b) wins, something genuinely new appears: the first review mechanism ever attached to provider-initiated scanning of private messages. A national judicial or independent administrative authority would see the search plan — the technologies, the safeguards. That is weak oversight. It is not targeting, and it does not exclude end-to-end encryption from anything. But it is the only mechanism on the table that requires anyone outside the provider to look at what the provider scans. If your community's messages live on a hosted platform, (a) versus (b) is the difference between never and rarely.

If (c) wins, the permanent law stops short of private messages — and scanning continues anyway under 2026/1881 until April 2028. Celebrate nothing. What (c) actually buys is time and a clean legal seam: after April 2028, the voluntary scanning of private messages would need a fresh affirmative act to continue, instead of a permanent permission nobody voted on. That is the real difference between a permanent regime and a lapsing one — each extension has to win a vote again.

The constant is unchanged either way: the machinery gets built. The EU Centre, the indicator database, the national coordinating authorities — technical agreement on the public-content piece is already recorded, and the seat process launches if Tuesday confirms it. Infrastructure outlives mandates. A regulation that excludes private messages from scanning still creates the database that would scan them if a later law asks.

And the architecture answer does not move: messages that never transit a provider's scanning surface are not covered by any of the three letters. Run the homeserver. Own the disk. The perimeter question is being negotiated in LIMITE documents; the only fixed point is the one you hold.

Tuesday's trilogue decides which letter survives. The dates that will actually tell you what scanning does: October 1, when the report-recipients list goes public, and February 1, when the error rates follow.