What do you do when the containment feature your Copilot pilot was built on stops accepting new tenants? Microsoft blocked new enablement of Restricted SharePoint Search on 31 July 2026, and its own documentation says the feature was never a security boundary.
You contain Copilot oversharing by fixing permissions on the sites Copilot can reach and by limiting the pilot to a small group of licence holders. Restricted SharePoint Search stopped accepting new enablement on 31 July 2026, per the [Microsoft Learn page](https://www.learn.microsoft.com/en-us/sharepoint/restricted-sharepoint-search) updated 9 October 2026. Budget around $30.00 per user per month and a quarter of remediation before you buy seats.
## Why the allowlist was always rented time
RSS let an admin restrict organisation-wide Copilot and search results to a curated list of SharePoint sites while the permissions underneath stayed exactly as broken as they were. Microsoft describes it plainly as a temporary measure that “isn’t a security boundary” and that does not change SharePoint permissions. Two details in that same documentation make the point sharper than any commentary could.
First, the allowlist tops out at 100 sites. Hub sites count against the cap; sites associated with a hub do not. For a Swiss SME with a tidy information architecture, 100 hubs is generous. For a 400-person organisation that has been growing SharePoint by department for a decade, it is a budget you spend arguing over.
Second, and this is the part that undermines the whole premise: even with the restriction active, users can still interact through Copilot with files they own or have previously accessed, per Microsoft’s Copilot Search FAQ. So if someone was wrongly given access to the HR compensation library in 2023 and opened it once, RSS does not take that away. It hides the site from org-wide discovery while leaving the actual exposure intact.
Restricted SharePoint Search contains Copilot oversharing while you fix permissions in the background.
Microsoft’s own documentation calls it a temporary measure that is not a security boundary, does not alter SharePoint permissions, and still allows Copilot interaction with files a user owns or previously accessed.
My read: RSS bought time, and plenty of organisations used that time well. The ones in trouble now are the ones who treated the allowlist as the deliverable rather than the scaffolding. If your 2025 Copilot risk assessment has a line that says “mitigated by Restricted SharePoint Search”, that line is no longer true for new tenants and was never fully true for anyone.
## A three-month delay on 150 seats costs about $13,500
The oversharing problem is not a niche Swiss concern. A 2026 Gartner-referenced survey cited in Opsin’s Microsoft Copilot Agents research found 51% of respondents named oversharing and data loss the top barrier to successful Copilot deployment. A separate Gartner-referenced figure reported by Medha Cloud puts roughly 40% of IT leaders in rollout delays of three months or more, with data oversharing as the main cause.
Put those two numbers against the licence price and the arithmetic gets uncomfortable. Microsoft 365 Copilot is listed at $30.00 per user per month paid yearly on the [Copilot Studio pricing page](https://www.microsoft.com/en-us/copilot/pricing/copilot-studio). Copilot Business currently shows as “originally starting from $21.00, now starting from $18.00” per user per month. A 150-seat annual commitment at $30 is $54,000 a year. If permission remediation then pushes go-live out by three months, you have paid roughly $13,500 for an entitlement nobody is allowed to use yet.
Forrester’s Total Economic Impact study for SMB Microsoft 365 Copilot projects a three-year ROI range of 132% at the low end, 243% medium, 353% high. Those are projections built on a composite, and I treat the spread itself as the useful signal: the gap between 132% and 353% is largely a function of how fast and how broadly people actually use the thing. I have written before about [how badly enterprise AI ROI numbers get quoted](https://www.arturmarkus.com/enterprise-ai-roi-5-numbers-everyone-quotes-wrong-from-95-fail-to-mckinseys-37/), and this one deserves the same caution: a projected range is a planning input, not a promise.
There is a line item people forget. Agents grounded only in their own instructions or public websites incur no extra charge. Agents that touch tenant data through SharePoint or connectors consume metered Copilot Credits, per Microsoft’s cost considerations documentation. Copilot Studio credit packs run $200 per month for 25,000 credits, billed annually, and there is also an Azure-linked pay-as-you-go meter billed monthly on credits actually consumed with no upfront capacity purchase. For an SME that cannot forecast agent usage, pay-as-you-go is the lower-regret starting point.
## How do you contain Copilot oversharing now that Restricted SharePoint Search is blocked?
You make the permission model correct for the sites Copilot is allowed to index, and you limit the pilot population rather than the search surface. The order below is the one I would follow, and the reasoning matters more than the sequence.
Settle data residency before you provision anything
Microsoft lists Switzerland among supported Local Region Geography sign-up countries for Copilot data residency, alongside the EU, Germany, France, Sweden, Norway and the UK. Advanced Data Residency requires both an eligible tenant geography and an ADR subscription covering all users in the tenant, and existing tenants may need a Global Admin opt-in migration. That migration is not something you want to discover halfway through a rollout, and “all users in the tenant” means the cost is not a pilot-sized cost.
Inventory what Copilot would see, not what you think exists
The point of this pass is to find the gap between your mental model of the SharePoint estate and the actual permission graph: broad “Everyone except external users” grants, orphaned sites, old sharing links. RSS let you skip this step. Nothing available now does.
Fix permissions at the source, starting with the highest-sensitivity sites
Microsoft is explicit that RSS never changed SharePoint permissions. That means every hour spent on the allowlist was an hour not spent on the underlying grants. Sequence by sensitivity (HR, finance, legal, board material) rather than by site size, because the damage from one wrongly surfaced compensation file outweighs a hundred stale project folders.
Pilot by population, not by index
With the index-level stopgap gone, the controllable variable is who holds a licence. A small cohort drawn from a department whose sites you have already remediated keeps blast radius proportional to the work completed. It also gives you real usage data before you commit to an annual seat count at $30 per user per month.
Decide agent grounding deliberately
Agents grounded in instructions or public websites cost nothing extra and carry no tenant-data exposure. The moment an agent reaches into SharePoint or a connector, it both bills credits and inherits whatever permission mess remains in that scope. Treat the grounding decision as a security decision that happens to have a price tag.
Map your Article 50 obligations onto the same timeline
EU AI Act Article 50 transparency obligations applied from 2 August 2026, and systems placed on the market before that date face a 2 December 2026 deadline for marking and detection, per the EU AI Act Service Desk FAQ. If your Copilot deployment includes customer-facing agents, the marking work belongs in the same programme as the permission work rather than in a separate compliance workstream that discovers the deadline in November.
The uncomfortable part of this sequence is that steps two and three have no fixed duration. The 40% figure on three-month delays is the only quantified signal I have on how long remediation takes in practice, and it describes organisations that were blocked, not a median. Unconfirmed: I would plan for a quarter of permission work on a mid-sized estate and treat anything faster as a pleasant surprise.
## Three ways this rollout usually goes wrong
The first is buying seats on the delivery date. Annual Copilot commitments start billing whether or not your permission remediation finished. With roughly 40% of IT leaders reporting delays of three months or more driven by oversharing, the overlap between “licences active” and “users blocked” is likely. **Stage the licence purchase behind the remediation milestone, not beside it.**
The second is treating ownership of the permission fix as an IT problem. The grants that cause trouble were usually created by business users with legitimate intent, and IT cannot adjudicate whether the finance controller should still see a 2022 acquisition folder. Remediation stalls when it arrives on an infrastructure team’s backlog without a named business owner per site collection.
The third is the Article 50 collision. Transparency obligations applied from 2 August 2026, with 2 December 2026 as the deadline for systems already on the market. A Copilot Studio agent that talks to customers is squarely in scope for marking, and the engineering effort is small while the discovery effort is not. I covered the mechanics of that work in a [separate guide to shipping Article 50 marking](https://www.arturmarkus.com/how-to-ship-article-50-marking-before-december-2-2026-a-ctos-guide-to-the-eu-transparency-code/); the point here is calendar collision, not technique.
## When waiting out the quarter beats rolling out
There is an honest case for not rolling out Copilot this quarter, and it has nothing to do with licensing.
If your SharePoint estate has never had a permission review and you cannot name an owner for each major site collection, the remediation is organisational change work with a technical component. Starting a Copilot programme on top of that converts a governance problem into a visible AI incident. The file was already over-shared; Copilot just makes it findable by anyone who asks a plain-language question.
If you need Advanced Data Residency and sit on an existing tenant, confirm the Global Admin opt-in migration path and the all-users subscription requirement before committing to a date. A residency requirement that only surfaces in a legal review three weeks before launch will feel like a compliance problem, but it is a schedule problem.
And if your use case is genuinely a narrow, bounded agent, say an internal assistant grounded only in its own instructions or public web content, you may not need tenant-wide Copilot licensing at all for the first version. That configuration incurs no extra charge and no tenant-data exposure, which makes it a cheap way to find out whether your people actually want the thing before you underwrite 150 seats.
The removal of Restricted SharePoint Search is a net positive for anyone who was going to do the permission work anyway, and a genuine setback for organisations who were using it as a permanent answer. Microsoft removed an option that produced false confidence. My expectation is that the 51% who name oversharing as the top barrier will split: the ones with clean enough estates will accelerate, and the ones without will discover that their Copilot project was a SharePoint governance project with a nicer name on the business case.
## What I would tell a 200-person Swiss firm this month
Do the inventory now, buy the licences later. **The permission graph in your tenant is a fact you can establish in weeks**, and the Copilot business case depends entirely on what that fact turns out to be. Switzerland is on Microsoft’s Local Region Geography list, so residency is solvable, but Advanced Data Residency needs the tenant-wide subscription and possibly a migration, which makes it a procurement conversation rather than a config toggle. Keep the first agent cohort grounded in instructions or public sources so it costs nothing in credits while you learn. And put 2 December 2026 on the same plan as the remediation milestones, because Article 50 marking for anything customer-facing does not care how your SharePoint cleanup is going.
If you are mid-rollout and your risk register still cites Restricted SharePoint Search as a control, that is worth an hour of someone’s attention this week, and I am happy to [talk it through](https://www.arturmarkus.com/contacts/) if a second opinion would help.
