Large organizations and governments keep requiring that their data, and the AI that works on it, stay under their own control, in what is often called private or sovereign deployment, so this demand persists through 2027.

67.5% stated confidence, Likely (55–80%). Starting point on the record’s first day: Likely (55–80%), 67.5 per cent, the average of five readings, all five of Likely. Attention level: For information. Trend: Steady.

The judgment as stated before the record began

The main assessment holds only while large organizations and governments insist on keeping their data, and the AI that works on it, under their own control. That insistence shows in government policy that treats AI capacity as a matter of national security, in rules on where sensitive data may be kept and who may reach it, and in wariness of depending on a foreign supplier that could restrict or cut off access. Canada and Germany, Cohere's two home markets once its merger with Aleph Alpha closes, launched a Sovereign Technology Alliance in February 2026, and in June 2026 the European Commission proposed a common way to assess how far cloud and AI services are under European control, for use in public purchasing. The assessment would weaken if buyers came to accept the promises of shared public clouds as enough, if governments cut the money they have set aside, or if rules on where data is kept were relaxed.

How the record started this assessment

On 15 October 2026, before the record began, a model read the 10 developments below five separate times, first reported in the three weeks before and linked to this assessment's causal factors or to its rivals', and chose a term each time. A second model, from a different family, checked each development a reading relied on, disputed 7 developments, and the five readings were made again without them. The record started the assessment at the average of the five final terms, taken in log-odds: 67.5 per cent, Likely (55–80%). Every reading and every check is kept in the first day's record.

The five final readings, in order: Likely, Likely, Likely, Likely, Likely. Their average, taken in log-odds, is 0.73, a stated confidence of 67.5 per cent. The readings were made by Claude Sonnet 4.6 (Anthropic) and checked by Gemini 4 Argon (Google).

Factors named above: R1 (Government policy and money for AI under national control), R2 (Rules on where data is kept and who can reach it), R3 (Investment in clouds and computing capacity kept within a country), R4 (Wariness of depending on a foreign supplier) and R5 (Large buyers accepting AI from shared public clouds).

The developments shown

1. 1. First reported 9 October 2026, 2 independent publishers. Linked to this assessment's own causal factors.
   “Germany drops binding framework for sovereign AI partnership with Mistral and SAP” · sovereigntymonitor.eu · 9 October 2026 · Open the report · archived copy
   “Digital Ministry clarifies: No withdrawal from AI project with SAP and Mistral” · heise.de · 14 October 2026 · Open the report · archived copy
   First round: relied on by 3 readings as support for the assessment and 2 readings as counting against it. Removed from the second round, because the second model disputed it.
   • Check of its use as support for the assessment: Disputed by the second model, so the second round was read without it, run on Gemini 4 Argon (Google).

     Its reason, as the model wrote it: Factor R1 supports the assessment when governments commit policies, budgets, or binding agreements to pay for or favour sovereign AI. Here, the reports state that Germany dropped a binding framework and clarified its letter of intent involved no procurement or funding commitments, leaving the companies to continue independently. Because this shows an absence or scaling back of government commitment rather than new policy or funding support, it does not bear on R1 in the supporting direction.

     Its ground: what the reports state is not a change in this causal factor.
     • Report 1, quoted: “Germany drops binding framework for sovereign AI partnership with Mistral and SAP”
     • Report 2, quoted: “Germany’s digital ministry said its letter of intent with SAP and Mistral was non-binding and involved no procurement or funding commitments, while noting the companies continued developing public-sector technology independently.”
   • Check of its use as counting against the assessment: Upheld by the second model, run on Gemini 4 Argon (Google).

     Its reason, as the model wrote it: The reports state that Germany dropped a binding framework for a sovereign AI partnership with Mistral and SAP, clarifying that the arrangement involved no procurement or funding commitments. Under factor R1, a reduction or absence of expected government funding and agreements for AI under national control weakens the factor and therefore counts against the assessment that demand for sovereign deployment persists.

     • Report 1, quoted: “Germany drops binding framework for sovereign AI partnership with Mistral and SAP”
     • Report 2, quoted: “Germany’s digital ministry said its letter of intent with SAP and Mistral was non-binding and involved no procurement or funding commitments”
2. 2. First reported 7 October 2026, 2 independent publishers. Linked to this assessment's own causal factors.
   “EU Begins Broader Enforcement of AI Act Rules for General-Purpose Models - Brussels Post” · brusselspost.com · 7 October 2026 · Open the report · archived copy
   “Commission starts enforcing AI Act rules and new transparency requirements on 2 August | Shaping Europe’s digital future” · digital-strategy.ec.europa.eu · 9 October 2026 · Open the report · archived copy
   First round: relied on by 5 readings as support for the assessment. Removed from the second round, because the second model disputed it.
   • Check of its use as support for the assessment: Disputed by the second model, so the second round was read without it, run on Gemini 4 Argon (Google).

     Its reason, as the model wrote it: Factor R2 covers rules on where data is kept and who can reach it, requiring sensitive data and the AI that handles it to stay in a given country or under the owner's control. The reports only state that the EU is enforcing transparency and risk-management requirements for developers of general-purpose AI models, not rules on data residency, data access, or private deployment.

     Its ground: what the reports state is not a change in this causal factor.
     • Report 1, quoted: “European regulators have begun enforcing new transparency and risk-management requirements for developers of powerful general-purpose artificial intelligence systems as member states expand oversight under the EU Artificial Intelligence Act.”
     • Report 2, quoted: “Commission starts enforcing AI Act rules and new transparency requirements on 2 August”
3. 3. First reported 2 October 2026, 2 independent publishers. Linked to this assessment's own causal factors.
   “Amazon and Microsoft's cloud units set to fall under EU's Digital Markets Act: report | Seeking Alpha” · seekingalpha.com · 2 October 2026 · Open the report · archived copy
   “EU Regulators Set to Designate AWS and Azure as Digital Markets Act Gatekeepers - NewAnalytica” · newanalytica.com · 4 October 2026 · Open the report · archived copy
   First round: relied on by 5 readings as support for the assessment. Removed from the second round, because the second model disputed it.
   • Check of its use as support for the assessment: Disputed by the second model, so the second round was read without it, run on Gemini 4 Argon (Google).

     Its reason, as the model wrote it: The reports state that designating AWS and Azure as gatekeepers under the EU Digital Markets Act concerns competition, pricing, data portability, interoperability, and switching between cloud providers. This does not constitute a rule on where sensitive data is kept or who can reach it, nor does it require data or AI to remain in a given country or under the owner's private/sovereign control (Factor R2). Nor does it represent buyers accepting shared public clouds for sensitive work (Factor R5).

     Its ground: what the reports state is not a change in this causal factor.
     • Report 2, quoted: “a gatekeeper designation would force both companies to enable easier data portability, interoperability, and switching between providers.”
4. 4. First reported 14 October 2026, 1 independent publisher. Linked to this assessment's own causal factors.
   “France carves its own path in AI race with Nvidia’s help” · euronews.com · 14 October 2026 · Open the report · archived copy
   First round: relied on by 5 readings as support for the assessment. Second round: relied on by 5 readings as support for the assessment.
   • Check of its use as support for the assessment: Upheld by the second model, run on Gemini 4 Argon (Google).

     Its reason, as the model wrote it: The report explicitly describes France's sovereign technology push, including a government deployment involving a French digital workplace and a domestic AI partner. This directly fits factor R1 (government policy and action favoring AI under national control) and supports the assessment that government demand for sovereign AI deployment persists, while going beyond mere funding as in alternative RX1.

     • Report 1, quoted: “France carves its own path in AI race with Nvidia’s help”
     • Report 1, quoted: “The report describes France’s sovereign technology push and notes a government deployment involving a French digital workplace, while identifying Mistral as a partner in the country’s broader AI ecosystem.”
5. 5. First reported 13 October 2026, 1 independent publisher. Linked to this assessment's own causal factors.
   “Government of Canada expands support to adopt and commercialize artificial intelligence” · cantechletter.com · 13 October 2026 · Open the report · archived copy
   First round: relied on by 5 readings as support for the assessment. Removed from the second round, because the second model disputed it.
   • Check of its use as support for the assessment: Disputed by the second model, so the second round was read without it, run on Gemini 4 Argon (Google).

     Its reason, as the model wrote it: The report describes the Canadian federal government making C$500 million in public funding available through a regional initiative as part of a broader national strategy. While this reflects government policy and funding under factor R1, it is an announcement of government funding and industrial policy rather than organizational purchases. It therefore fits alternative RX1—that demand is created by public money and industrial policy—just as well as it fits the assessment, and cannot separate the two.

     Its ground: it fits one of the rival explanations as well as the assessment.
     • Report 1, quoted: “Government of Canada expands support to adopt and commercialize artificial intelligence”
     • Report 1, quoted: “The federal government made an additional C$500 million available over five years through the Regional Artificial Intelligence Initiative, part of a broader strategy that also references secure computing infrastructure.”
6. 6. First reported 12 October 2026, 1 independent publisher. Linked to this assessment's own causal factors.
   “Decreto AI Act in vigore: obblighi, sanzioni e scadenze” · pmi.it · 12 October 2026 · Open the report · archived copy
   First round: relied on by 2 readings as support for the assessment. Removed from the second round, because the second model disputed it.
   • Check of its use as support for the assessment: Disputed by the second model, so the second round was read without it, run on Gemini 4 Argon (Google).

     Its reason, as the model wrote it: Factor R2 covers rules on where data is kept and who can reach it, requiring sensitive data and AI to stay in a given country or under the owner's control. The report only states that Italy's AI Act decree sets 'oversight roles, penalties, workplace requirements and near-term compliance dates,' with no mention of data localisation, residency, or buyer control over where data and AI are hosted.

     Its ground: what the reports state is not a change in this causal factor.
     • Report 1, quoted: “Italy published a national AI Act implementation decree setting oversight roles, penalties, workplace requirements and near-term compliance dates, while some EU high-risk AI obligations remain deferred.”
7. 7. First reported 11 October 2026, 1 independent publisher. Linked to this assessment's own causal factors.
   “European Commission launches public consultation on draft implementing act under ‘AI Act’ | AGENCE EUROPE” · agenceurope.eu · 11 October 2026 · Open the report · archived copy
   First round: relied on by 2 readings as support for the assessment. Removed from the second round, because the second model disputed it.
   • Check of its use as support for the assessment: Disputed by the second model, so the second round was read without it, run on Gemini 4 Argon (Google).

     Its reason, as the model wrote it: The reading claims that this development strengthens data-localisation and transparency obligations that push large buyers toward AI they control. However, Report 1 only states that the European Commission opened a public consultation on a draft regulation concerning 'practical arrangements for the penalty procedures' under the AI Act. The report does not state anything about data localisation, where data is kept, or keeping data and AI under a buyer's own control.

     Its ground: the reports do not state the development the reading relies on.
     • Report 1, quoted: “On Monday 12 October, the European Commission opened a public consultation, running until 9 November, on a draft regulation setting out the practical arrangements for the penalty procedures under the Artificial Intel”
8. 8. First reported 9 October 2026, 1 independent publisher. Linked to this assessment's own causal factors.
   “AI Act Implementation and Adoption Timeline” · ibm.com · 9 October 2026 · Open the report · archived copy
   First round: relied on by 3 readings as support for the assessment. Removed from the second round, because the second model disputed it.
   • Check of its use as support for the assessment: Disputed by the second model, so the second round was read without it, run on Gemini 4 Argon (Google).

     Its reason, as the model wrote it: The reading relies on this development showing that the EU AI Act strengthens data-localisation obligations that push buyers toward AI they control. However, Report 1 only describes an IBM overview of general AI Act implementation milestones for general-purpose AI models and high-risk systems, and does not state anything about data-localisation rules, where data must be kept, or keeping AI under the owner's control.

     Its ground: the reports do not state the development the reading relies on.
     • Report 1, quoted: “IBM outlines the AI Act’s implementation milestones, including the general application of rules in August 2026 and later deadlines for legacy general-purpose AI models and certain high-risk systems.”
9. 9. First reported 9 October 2026, 1 independent publisher. Linked to this assessment's own causal factors.
   “While the world watches Microsoft and Anthropic, Canada is having its AI moment” · theglobeandmail.com · 9 October 2026 · Open the report · archived copy
   First round: relied on by no reading. Second round: relied on by 5 readings as support for the assessment.
   No reading relied on it, so it was not checked.
10. 10. First reported 8 October 2026, 1 independent publisher. Linked to this assessment's own causal factors.
    “Valcon launches highly secure, on-premise, air-gapped data platform, partnering with Databricks to accelerate AI adoption in sovereign environments” · valcon.com · 8 October 2026 · Open the report · archived copy
    First round: relied on by no reading. Second round: relied on by 5 readings as support for the assessment.
    No reading relied on it, so it was not checked.

The readings, and their reasons

Readings made before this one

No starting point was read before this one, as the first day records.

Causes of uncertainty

• Sovereign AI is a political aim as well as a buying requirement, and some announcements are statements of intent with no purchase behind them.
• Demand for control is strongest in government, defence and regulated industries and weaker elsewhere, so it can look broad while resting on a few kinds of buyer.
• Regions of shared public clouds kept inside one country, with encryption keys held by the customer, may satisfy some buyers' rules, which blurs the line between control and a promise of control.

What would change the judgment

• Governments in Cohere's main markets cutting the money they have set aside for AI under national control, or withdrawing proposed rules that favour it.
• Rules that keep sensitive government data inside the country being relaxed.
• Large banks, insurers and health systems moving sensitive AI work to shared public clouds.

The causal factors, and which way each one cuts

What the assessment assumes, and what would break it

• RA1. Governments in Cohere's main markets keep paying for and favouring AI under national control, in Canada through its sovereign compute strategy and in the European Union through its proposed sovereignty rules.
  If this fails, the assessment fails.

  Invalidated if, before the end of 2027, the Government of Canada ends the Canadian Sovereign AI Compute Strategy or moves most of its remaining money to other purposes, as stated in a federal budget, an economic statement, a news release on canada.ca or the strategy's own page, or the European Commission withdraws its proposal for the Cloud and AI Development Act (procedure 2026/0138(COD)), as recorded in the European Parliament's Legislative Observatory. Moving money between parts of the Canadian strategy does not count.

  Source: Innovation, Science and Economic Development Canada, Canadian Sovereign AI Compute Strategy, European Parliament, Legislative Observatory, procedure file 2026/0138(COD), Cloud and AI Development Act. The criterion’s fingerprint 8be6a344…856a was fixed on 15 October 2026 and is part of the record’s first day.
• RA2. The largest cloud companies keep selling services that hold customers' data and AI within a country or on the customers' own premises, because buyers keep asking for them.
  If this fails, the assessment weakens.

  Invalidated if, before the end of 2027, two or more of Amazon Web Services, Google and Microsoft announce, on their own blogs or news sites, that they will close or stop selling a service that keeps customers' data and AI within a country or on the customers' own premises, such as the European Sovereign Cloud of Amazon Web Services, Gemini on Google Distributed Cloud, or Microsoft's sovereign private cloud.

  Source: Amazon Web Services, Opening the AWS European Sovereign Cloud, 14 January 2026, Google Cloud, Gemini is now available anywhere, 28 August 2025, Microsoft, Microsoft Sovereign Cloud adds governance, productivity and support for large AI models securely running even when completely disconnected, 24 February 2026. The criterion’s fingerprint 88e9d259…85e2 was fixed on 15 October 2026 and is part of the record’s first day.

The rival explanations

Evidence moves the assessment only when it bears differently on the assessment than on at least one of these. Evidence that fits them equally carries no weight.

• RX1. Demand for AI under national control is created by public money and industrial policy, and would fade if that support were withdrawn.
  No development was recorded on this assessment in the last 7 days.
  Favoured by:
  • Announcements of sovereign AI consist mostly of government funding rather than purchases by organizations.
  • Private companies choose shared public clouds even where government policy favours national control.
  Ruled out by:
  • Private organizations without public funding buy private deployments in growing numbers.
• RX2. Buyers settle for promises inside the largest clouds: control comes to mean a region of a US provider's cloud kept inside the country, rather than AI run under the buyer's own control.
  No development was recorded on this assessment in the last 7 days.
  Favoured by:
  • Governments clear regions of US-owned clouds kept inside the country for their most sensitive data.
  • Public tenders that once required hosting in the country or on the buyer's own premises accept those regions.
  Ruled out by:
  • Tenders and rules keep requiring infrastructure that is owned or operated in the country, or on the buyer's premises.

A development is recorded when the rule applies it to the assessment or sets it aside. Each count comes from the readings sealed with the changes, and a development set aside as fitting every alternative equally tells no alternative apart.

Evidence behind changes in the last 30 days

Each item is a report or data series cited by a sealed change on this assessment. The update loop reads headlines and summaries, and no summary is shown as the evidence.

No evidence has been cited in the last 30 days.

Evidence considered and set aside

Developments the update loop read in the last 30 days and set aside because they fitted this assessment and each of its rival explanations equally. Under the rule such evidence carries no weight, so none of it moved anything.

Nothing on this assessment was set aside in the last 30 days.

Changes to this assessment

No change to this assessment yet.

The rest of the judgment

• Main assessment
• The alternative

The question’s text on this page is as sealed on the record’s first day. Fingerprint 412510ac…302b. Fingerprint pending. Records are fingerprinted each day at 23:59 UTC.