The numbers are stated judgments. The record began on 15 October 2026 with starting confidences that a model set as the average of five separate readings, from its general knowledge and up to ten developments for each assessment first reported in the three weeks before, after a second model had checked the developments the readings relied on, and a published rule has made every change since. They are not measured probabilities. The question and the rule are fixed before the record begins, and each change shows its evidence and its reasons. The rule and what is not claimed are on the method page.
Through 2027, Cohere keeps growing by selling secure, private AI to large organizations and governments, rather than by competing to build the most capable models.
Status: Holds The alternative is the main assessment's rival: the two cannot both be right. The thesis holds when the main assessment stands at 60 per cent or more, the condition it rests on at 50 per cent or more, the alternative below 50 per cent, and no recent change has moved toward the alternative. It is inverted when the main assessment falls below 40 per cent, when the alternative is more likely than the main assessment, or when several recent changes have moved toward the alternative. Otherwise it is strained.
As of 15 October 2026, 10:48 UTC, day 0 of the record, which began on 15 October 2026.
Last checked for news: 15 October 2026, 12:31 UTC. The update loop checks once a day, and each day’s state is written at the end of the day.
What to expect. The rule moves a number only after at least 2 independent publishers report the same development within 7 days of its first report. One development then moves an assessment by at most about 2.2 percentage points when an outlet graded high reports it, and by under one percentage point when every outlet reporting it is graded unknown. Of the 26 outlets the update loop has read for this record so far, 18 are graded unknown. Of the 4 developments new to the record since it began that the update loop has read, 0 had a second independent publisher when it read them. At these rates changes are expected to be rare.
What the thesis rests on, and what would replace it
Through the end of 2027, how does Cohere grow: by selling secure, private AI to large organizations and governments, or by competing to build the most capable models?
Cohere's sales of secure, private AI to large organizations and governments keep growing through 2027 and remain its main source of growth.
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.
Larger rivals, among them the largest US AI and cloud companies and the French model maker Mistral AI, offer equivalent private and sovereign deployments and win the customers Cohere targets, so by the end of 2027 Cohere's sales of secure, private AI to large organizations and governments stop growing.
What the update loop read
Each daily state records what the update loop read in its day. Set aside means a development fitted the assessment and each of its rival explanations equally, so it carries no weight.
The reading is recorded from the record’s second day.
What the update loop read, and what the rule made of it
These are the update loop's working notes: each development it read, how it read it, and what the published rule makes of that reading. They are not part of the sealed record and they move nothing. Only a sealed change, made by the rule, moves a number.
Since the record began, the update loop has read 4 developments first reported after it began, in 5 readings, one for each assessment a development bears on. 5 readings are waiting for a second outlet, and 0 were set aside.
The daily states record what the update loop read from the record's second day, and none has been sealed yet. Both count a development once for each assessment it bears on. The daily states also count reports, while these notes list developments, and a report retrieved after the record began but first reported before it is counted in the daily states and never weighed by the rule.
Each outcome is worked out by the published rule from what the loop recorded when it read the development: its outlets then, its reading, and the challenge step's verdict once that has run. The outlet grades are the loop's own.
- Read 15 October 2026, 12:31 UTC · The condition it rests on
- “AI, resources and water bombers: Draft document maps Canada, EU path to closer ties” · kitchener.citynews.ca · 15 October 2026 · Open the report · archived copy
First reported 15 October 2026. Reported by 1 outlet when it was read. Outlet grade: unknown.Strengthens factor R3 (Investment in clouds and computing capacity kept within a country). Read as routine.- On the assessment: more likely.
- On alternative RX1: neither more nor less likely. This tells the alternative apart from the assessment.
- On alternative RX2: less likely. This tells the alternative apart from the assessment.
Its reason, written by Claude Sonnet 4.6 (Anthropic): The draft Canada–EU statement links digital sovereignty to cross-border infrastructure cooperation, signalling that both governments are actively building frameworks to keep cloud and AI capacity under national or allied control rather than relying solely on US hyperscalers. This strengthens R3 (investment in sovereign computing capacity) and cuts against RX2 (settling for regions of US-owned clouds), though as a draft document it is commentary rather than a binding contract or regulatory order.
Outcome: Waiting for a second independent publisher to report it, until 22 October 2026. - Read 15 October 2026, 12:31 UTC · The alternative
- “TPIsoftware Launches New Series of Sovereign AI Solutions Designed for Legal, HR, Education and Enterprise Governance” · us.acrofan.com · 15 October 2026 · Open the report · archived copy
First reported 15 October 2026. Reported by 1 outlet when it was read. Outlet grade: unknown.Strengthens factor C1 (Private and sovereign offerings from larger rivals). Read as routine.- On the assessment: more likely.
- On alternative CX1: neither more nor less likely. This tells the alternative apart from the assessment.
- On alternative CX2: neither more nor less likely. This tells the alternative apart from the assessment.
- On alternative CX3: neither more nor less likely. This tells the alternative apart from the assessment.
Its reason, written by Claude Sonnet 4.6 (Anthropic): TPIsoftware, a rival vendor, is launching sovereign AI solutions with on-premises inference and fine-tuning across multiple enterprise sectors, adding another competitor in the private/sovereign AI space that Cohere targets. However, TPIsoftware is not one of the named larger rivals (major US cloud companies or Mistral AI), so the effect on C1 is marginal and the development is commentary-level rather than a signed contract displacing Cohere.
Outcome: Waiting for a second independent publisher to report it, until 22 October 2026. - Read 15 October 2026, 12:31 UTC · The alternative
- “Etisalat UAE, DataRobot extend sovereign agentic AI deal” · middleeastainews.com · 15 October 2026 · Open the report · archived copy
First reported 15 October 2026. Reported by 1 outlet when it was read. Outlet grade: unknown.Strengthens factor C1 (Private and sovereign offerings from larger rivals). Read as material.- On the assessment: more likely.
- On alternative CX1: neither more nor less likely. This tells the alternative apart from the assessment.
- On alternative CX2: neither more nor less likely. This tells the alternative apart from the assessment.
- On alternative CX3: neither more nor less likely. This tells the alternative apart from the assessment.
Its reason, written by Claude Sonnet 4.6 (Anthropic): A larger rival (DataRobot) signs a multi-year on-premises sovereign AI deal with a major Gulf telco, demonstrating that competitors are actively winning the kind of secure, private enterprise deployments Cohere targets, which strengthens the case that rivals can displace Cohere in this segment.
Outcome: Waiting for a second independent publisher to report it, until 22 October 2026. - Read 15 October 2026, 12:31 UTC · The condition it rests on
- “Etisalat UAE, DataRobot extend sovereign agentic AI deal” · middleeastainews.com · 15 October 2026 · Open the report · archived copy
First reported 15 October 2026. Reported by 1 outlet when it was read. Outlet grade: unknown.Strengthens factor R3 (Investment in clouds and computing capacity kept within a country). Read as material.- On the assessment: more likely.
- On alternative RX1: less likely. This tells the alternative apart from the assessment.
- On alternative RX2: less likely. This tells the alternative apart from the assessment.
Its reason, written by Claude Sonnet 4.6 (Anthropic): The deal is explicitly on-premises and locally controlled, confirming that large organizations in the Gulf are still demanding sovereign, under-their-own-control AI deployments rather than settling for US-cloud regions or relying solely on government subsidies, which supports the persistence of demand that the thesis rests on.
Outcome: Waiting for a second independent publisher to report it, until 22 October 2026. - Read 15 October 2026, 12:31 UTC · Main assessment
- “Cohere to Acquire AISLE for Cyber-Resilient Sovereign AI” · cohere.com · 15 October 2026 · Open the report · archived copy
First reported 15 October 2026. Reported by 1 outlet when it was read. Outlet grade: medium.Strengthens factor M3 (Partners that bring Cohere's models into private settings). Read as material.- On the assessment: more likely.
- On alternative MX1: less likely. This tells the alternative apart from the assessment.
- On alternative MX2: more likely.
- On alternative MXC: less likely. This tells the alternative apart from the assessment.
Its reason, written by Claude Sonnet 4.6 (Anthropic): Cohere's proposed acquisition of AISLE is a concrete step to deepen its secure, sovereign AI capabilities for governments and regulated organizations across private-cloud, on-premises, and air-gapped environments, directly widening routes to private deployments and strengthening M3. This is a material deal that supports the main assessment while cutting against MX1 (capability race) and MXC (rivals winning the space), though it does partially support MX2 since growth here comes via acquisition rather than organic sales.
Outcome: Waiting for a second independent publisher to report it, until 22 October 2026.
All the working notes, including 19 readings from before it began.
The latest changes
No changes yet.
The changes page lists every change (none so far), including moves the rule set aside and any the daily check refused. The weekly brief puts each week of the record on one printable page.
Every daily state and every change is fingerprinted
Each record’s fingerprint (its SHA-256 hash) joins that day’s fingerprint of all such records, and the day’s fingerprint is published in a public repository on GitHub. Your browser can recompute a record’s fingerprint and compare it with the published one.
The first day (15 October 2026), with the starting point and the rule, which is also the latest daily state: Fingerprint . Fingerprint pending. Records are fingerprinted each day at 23:59 UTC.
Choose a record to check. The state and the changes are also published as data: the state and the changes, as JSON Lines (empty until the first change), and an Atom feed of both, to follow the record in a feed reader.
To run a thesis of your own on this machinery, write to cberzins@uottawa.ca.