Research & Commentary
The Language Firm's published research record.
Where the firm's recurring intelligence is lifted into argument, generalized beyond its original cases, and addressed to the readers positioned to act on it at scale.
The Language Firm's standing body of analytical work.
-
![]()
Audit Reports & Policy Briefs
The First Watch series, released on the 6th of every month, is one rotation cycle of the Drift Audit, conducted under The Forensic Read™ against a fixed roster of K-12 AI tools in widespread district use. The audits are the firm's longitudinal record of vendor governance drift. Policy briefs are the cross-cycle synthesis, addressed to federal, state, and district leaders: where an audit reports what one rotation found, a brief reports what multiple rotations mean. Governance at the Speed of Drift is the first.
-
![]()
Incident Memos
Reads the prior month's major K-12 vendor breaches after the month has closed, when the record is complete. It does not rank the breach or re-report it. It uses the breach as the way into the governing documents underneath: it reads the vendor's actual terms against what a district believes it agreed to, and surfaces the governance gap the breach exposes, where student data is left undefined, quietly reclassified, or moved outside the district's agreement altogether, and where no one on the district's side is named to account for it.
-
![]()
White Papers
Standalone arguments on the questions the audit record surfaces but does not, on its own, resolve. White papers carry the methodology, the structural findings, and the cross-sector portability arguments in the form that researchers, agencies, and journalists can cite directly. Published as the underlying research lines mature.
Audit Reports & Policy Briefs
The First Watch: A Drift+The Forensic Read™ Report
-
The First Watch is a longitudinal record of how K-12 AI tools move underneath their own marketing surfaces between procurement reviews.
-
Every audit takes a fixed roster of tools already in widespread district use, places each one against a dated baseline, and reports what changed at the policy layer, the enforcement layer, and the product surface during the rotation window. Each finding is cross-referenced against the vendor's own primary sources, including privacy policies, terms of service, data processing agreements, subprocessor disclosures, admin-console release notes, and the vendor's own announcement channels. No third-party summaries, no AI-generated commentary, no speculation. One tool is added to the roster per month.
-
The Drift Category: No Drift, Watch, or Flag. This describes whether the existing Tool Spotlight Card on file still describes the tool accurately, whether new context warrants practitioner awareness, or whether the assessment requires update before continued use.
The Forensic Pattern: Convergence, Asymmetric Movement, or Divergence. This describes how the vendor's policy and posture moved relative to each other during the rotation window. Convergence means the vendor said and did the same thing. Asymmetric means one layer moved while another held still. Divergence means the vendor's posture and its policy moved in opposite directions.
The Tool-by-Tool Findings: for each tool, the dated baseline, the dated movements observed during the rotation, and the citation trail back to the vendor's own published source.
The Forensic View: the cross-tool reading. What the rotation as a whole reveals about how the K-12 AI vendor market is moving this cycle.
What the Practice Requires: the operational steps a district can take this week to stand up the same verification cadence on its own roster.
Incident Memos
-
May 2026 was defined by the Instructure / Canvas data exposure. This memo does not rank or re-report the breach. It uses it to surface the governance gap underneath: the moment a teacher signs a tool up alone, the same student record can fall under a different contractual regime, or outside the district's agreement entirely.
-
This companion instrument accompanies the first Incident Memo so each factual claim can be read against its source. The governing-document quotations were confirmed verbatim against the vendors' published terms, and the incident facts were traced to Instructure's own incident-update page where available and to multiple independent contemporaneous reports otherwise, fourteen sources in all. Each claim is classified by how it was verified, and where the memo's wording diverged from the record, the divergence is stated in full and its correction shown, not resolved silently. A claim is only as good as the document behind it.
-
A companion read to the May memo. The report read the gap that opens at the district's end, a teacher's solo signup. This takes the same Canvas breach as a test of the agreement that authorized the vendor, and reads the gap that opens at the vendor's end: what the breach-notification clause required, how the vendor performed against it, and how it treated districts afterward.