What shipped in the platform.
Release notes for the ArthurAI™ platform — the product itself, not this website. One page, because the 6 editions (VLE · CLE · SLE · ULE · FLE · PLE) are configurations of one platform: one reasoning layer, one component kit, one Arthur, one deploy. Where a change reached a single edition, the entry says so. ArthurAI™ Studio is a separate product on its own cadence and has its own section below.
Security, access-control, authentication and student personal-data changes are not itemised publicly. On a platform used by minors, a dated public list of which control changed and when is an attack map rather than a release note — and in this window that withheld work is the larger share of the engineering effort, not a rounding error. Customers receive the full detail through their engagement; our security posture and disclosure process are documented in the Trust Center.
One platform, six configurations.
- August 2026
Ask a teacher, from inside the conversation
A learner working with Arthur can escalate the conversation to the teachers who own the course, carrying the transcript with it. The tutor no longer has to be the end of the line when a learner is stuck — the educator is the decider, so reaching one is a first-class action rather than an instruction to leave the product and send an email.
- August 2026
Arthur’s reasoning stays with the transcript
The glass-box reasoning summary is now recorded against each of Arthur’s messages and rendered when the transcript is read back, not only while the answer is streaming. An educator reviewing a session after the fact sees the reasoning the learner saw. Saved study notes carry the same provenance: what they came from is recorded on the note and shown on the note.
- August 2026
Tell Arthur when an answer missed
Per-message feedback and a regenerate action on Arthur’s answers. Regenerate genuinely re-answers rather than replaying the same text, and feedback is suppressed on messages that have not yet been persisted — voting on a message that is about to be replaced only teaches the learner that the control does nothing.
- August 2026
Learning-native actions on an answer
Save to lesson notes, save to deck, quiz me on this, copy and share. An answer worth keeping can be turned into study material in one action instead of being copied out by hand.
- August 2026
An honest wait
Lesson generation used to narrate invented progress steps while a learner waited. It now shows real elapsed time and the actual state, backed by a platform generation-health signal — and a lesson is marked generated only when every step in it actually has content, so a lesson can no longer present itself as ready and then open empty.
- August 2026
Arthur confirms it is AI when asked
Asked directly, the tutor states plainly that it is AI, on every surface, without naming any underlying model — and generated content carries a disclosure in the interface. This is published as posture, not as a fix: a learner is entitled to a straight answer to that question.
- August 2026
Arthur’s register follows the learner
Tone and vocabulary flex by age band, so the same platform can address a secondary-school learner and a graduate cohort without one of them being talked down to. Arthur can also give directions: it answers "where do I find…" about the product itself, grounded in the help material rather than guessed.
- August 2026
A guided tour that exists, and can be replayed
A first-run tour for a new account, and — the part that was missing — a way to start it again later from the help panel. An orientation you get once, on the one day you are least able to use it, is not orientation.
- August 2026
Mathematics renders as mathematics
Arthur’s formatting was mixing notation into prose, so arrows and symbols in ordinary sentences rendered as raw markup while real equations sometimes did not render at all. Prose now uses ordinary symbols and only genuine mathematics is typeset — on the institutional Console and on PLE alike.
- August 2026
Practice questions fail loudly instead of stalling
When part of a practice set could not be produced, the whole step used to wait indefinitely. Valid questions are now kept and served, and a genuine failure surfaces as a failure — a learner waiting on nothing is worse than a learner told to try again.
- August 2026
Progress and dashboards read the real numbers
The learner dashboard was deriving progress from a stale count rather than the lesson record. It now reads the real one. Alongside it: empty states in the Console library and progress views say what is actually empty instead of implying content that was never there, and chart score ranges are clamped so an empty chart no longer advertises a range it has no data for.
- August 2026
Notifications reach people
Notification writes had been failing on every event since launch, so a whole class of notification silently never arrived; and a new-student notification rendered with the name missing. Both are fixed. Application telemetry also stopped being sampled, so a rare failure is visible rather than averaged away.
- August 2026
Reports produce correct figures, and recover
Money columns are converted correctly in the report builders — a precision type reaching a numeric formatter produced wrong totals rather than an error — faculty report queries that were never implemented now are, snapshots paginate, and a report schedule that stalls is recovered instead of staying stuck.
- August 2026
Mobile: the conversation is usable on a phone
A keyboard-aware immersive chat shell, a composer that does not trigger zoom on iOS, wider message rows, and export and action menus that stay on screen. Learners on the household editions are mostly on phones; the surface now assumes that rather than tolerating it.
- July 2026
Glass-box tutoring on the learner surface
Arthur’s reasoning and its real tool calls stream to the learner as they happen, with a visible ledger of what it did — and the answer cites the lesson material it was grounded in. Agentic, not autonomous: the learner watches the work, and the educator remains the decider.
- July 2026
Studio-authored courseware, taught as authored
Courses authored in ArthurAI™ Studio import into an edition at full fidelity with their provenance verified, all question types included, and a per-course teaching mode that decides whether Arthur teaches the authored material verbatim or adapts it. Import is available across the editions, FLE included.
- July 2026
Per-institution and per-learner AI capability controls
An institution can turn individual AI capabilities on or off, and a learner-level override sits underneath it. Existing learners were unaffected by default at rollout — a control that changes behaviour the moment it ships is not a control, it is a migration.
- July 2026
Family accounts — FLE
Guardian accounts with delegated switching between children, per-child progress, per-child parental controls, and a parent PIN. Built for the home-schooling edition, where the buyer, the supervisor and the learner are three different people in the same house.
- July 2026
Certificates: correct, complete, and exportable in bulk
Certificates stopped being issued for courses that were not finished, and learners already affected were reconciled — including their underlying progress, not just the certificate status. Program certificates now carry the course list and take their date from the last course actually completed. Administrators can export every issued certificate as one archive, delivered by email as a link with readable filenames.
- July 2026
Stuck learners get unstuck
Concurrent writes to lesson steps could lock a learner out of their own progression. Writes are serialised, a failed lesson can be regenerated by the learner or in bulk by an administrator, enrolled courses are immutable to regeneration so a course cannot change under someone mid-way, and a lesson that is impossible to start no longer reports itself as simply "not started".
- July 2026
Smaller things that were quietly wrong
Real profile-picture upload where there had been a stub; learners can save their own settings; international phone numbers validate everywhere rather than only in one format; announcements gained an urgent tier; and on PLE the reading theme both changes the reading surface and follows the learner across devices.
The authoring environment ships on its own clock.
ArthurAI™ Studio is where courseware is authored, reviewed and delivered — to an ArthurAI edition, to the learning-management system an institution already runs, or as a verifiable credential. It is not a seventh edition, and it does not deploy with the editions, so its entries are kept separate rather than merged into the list above. What Studio does.
Security, access-control, authentication and student personal-data changes are not itemised publicly. On a platform used by minors, a dated public list of which control changed and when is an attack map rather than a release note — and in this window that withheld work is the larger share of the engineering effort, not a rounding error. Customers receive the full detail through their engagement; our security posture and disclosure process are documented in the Trust Center.
- August 2026
The Simulated Learner — a test suite for courseware
A course can be run against simulated learners before a real one sees it, and the findings come back as a list an author can act on. Test drive lets an author select findings and apply the fixes in a single revision rather than one edit at a time.
- August 2026
Course Health — pedagogical lint
Studio reviews a course the way a linter reviews code: objectives that no assessment tests, assessments with nothing teaching them, uneven load across modules. Findings are advisory and attributed, so an author can disagree with one.
- August 2026
Living Courses — freshness, and a course that ages
Published courses are scanned for staleness and carry an aging notice, so a course that was accurate when it was written does not keep presenting itself as current two years later.
- August 2026
Arthur Noticed — proactive workspace review
Arthur reviews the workspace unprompted and raises what it found — an unfinished course, a gap in a path, an objective with no assessment — instead of waiting to be asked a question the author does not know to ask.
- August 2026
Workspace Skills Graph
A portfolio view of what the workspace’s courses actually teach, and proposed courses for the gaps between them.
- August 2026
Chat with Arthur in the workspace
Glass-box and workspace-scoped: Arthur answers about the courses in front of it, shows its reasoning and its tool calls, and its answers carry actions — copy, share, or author the answer into a course. It also acts as a wayfinder, answering product how-to questions from the help library rather than improvising them.
- August 2026
Help that is part of the product
A deep help library, a getting-started checklist, guided tours, escalation to Arthur from any help article, and support requests raised from inside Studio rather than from an email client. The library was then brought to parity with the shipped product — help describing a version of the product that no longer exists is worse than no help.
- August 2026
Seatless reviewer links
A subject-matter expert can review a course and leave feedback without being given a seat, and their feedback can be applied in one action. Review is the step that most often waits on procurement; it no longer has to.
- August 2026
One-tap localisation
A translated copy of a course, created as its own course so the original is untouched and the translation can be edited on its own terms.
- August 2026
Weekly digest for workspace owners
An owner gets a weekly summary of what moved in the workspace by email. Listed here rather than in the July notes because the digest and the recertification stamps behind it did not actually send when they were first built — the entry belongs to the date they worked.
- August 2026
Exam integrity for regulated training
Assessment controls for training that has to stand up to an auditor, with attempt history retained as an append-only record.
- August 2026
Studio on tablets and phones
Collapsible outline, a scrollable toolbar, a wrapping header, menus that appear on touch devices, and a chat layout that holds together on an iPad. Verified across a device fleet in the browser engines customers actually use, not only in one.
- August 2026
Corrections worth naming
A first course that failed to generate used to report success. Onboarding state was shared where it should have been per-account, and empty states claimed content that did not exist. A course can now have only one live publication at a time, and restoring an earlier version no longer discards work. Published courses and learner progress are protected from an edit made upstream of them.
- July 2026
Studio v2.0 — author full courseware, deliver anywhere
Studio became a complete authoring environment: full courseware from a brief, an editable and agentic workspace where an author can revise, ask Arthur, or edit inline, authored courses as first-class objects in a library, and per-edition authoring for CLE, SLE, ULE, VLE and FLE.
- July 2026
Deliver anywhere: SCORM, xAPI, credentials, or straight into an edition
Real SCORM 1.2 and xAPI export as an importable package, verifiable credentials, export targeted at a specific ArthurAI edition, and a direct push into a live edition. A malformed package degrades with an explanation instead of failing opaquely.
- July 2026
Marketplace, fork and remix
Courses can be published, discovered and imported across organisations, and any course can be forked and adapted to another edition — a vocational course reworked for a university cohort without starting again.
- July 2026
Ground a course in your own material
Ingest your own sources and author against them, interview a subject-matter expert and let Arthur turn the conversation into a course, and close the loop with learner signal — carried without personal data — so a course improves from how it actually taught.
- July 2026
Glass-box authoring, and a maker’s-tool identity
A live reasoning stream and a visible tool-call ledger while Arthur works, and a dark/light interface built for someone making something rather than for a dashboard. Studio also became usable on a phone, with the co-author panel as a drawer.
A short list is not a quiet month.
Two rules govern this page, and both of them shorten it. Nothing appears until a customer can use it — capability sitting behind a switch that is off does not get an entry, however much work it took. And security, access-control, authentication and student personal-data changes are not itemised here at all, because a dated public index of which control changed when is an attack map. In the window above, that withheld work is the larger share. Customers see all of it through their engagement.