What it is (researched: unified May 2026)
Data Security Posture Management discovers, protects and investigates sensitive-data risk across the estate; DSPM for AI extends that to AI apps and agents. As of May 2026 the two merged into a single GA solution with unified visibility across traditional and AI environments, spanning Microsoft 365, Azure, Fabric and integrated third-party SaaS. What it surfaces:
- Posture assessments — oversharing exposure (joining SharePoint/SAM data to AI risk), label/DLP coverage gaps, risky configuration — as recommendations with affected-resource lists.
- AI observability — an inventory of AI apps and agents (including Agent 365) with recent activity, how many are high-risk, and how many have sensitive interactions.
- Asset explorer — a unified view of unlabelled/classified data by workload.
- Activity analytics — sensitive info types appearing in prompts/responses, including third-party/consumer AI reached from managed endpoints (the BYO-AI shadow inventory).
Operate it like a vulnerability programme
The mistake is treating DSPM as a poster. Run it as a scanner:
- Baseline at the start of any Copilot/AI rollout — the first assessment IS the readiness report's spine (the Copilot curriculum's stage 0).
- Triage cadence — new findings weekly to named owners; 'the dashboard is red' only helps if red pages someone.
- SLA by class — oversharing rides the SharePoint runbook's clock; sensitive-prompt spikes ride security review; coverage gaps ride the labeling backlog.
- Trend, not snapshot — exposure shrinking wave over wave is the exec artifact.
Honest limits (the serv365 house style)
It watches the Microsoft-visible world: unmanaged devices and network-invisible AI use escape it (endpoint coverage decides reach); findings lag on the assessment cadence; and it measures DATA posture, not agent BEHAVIOUR (run histories and action telemetry are the agent instrument — the Copilot curriculum). The gap list is part of the report.
What to watch (proofs)
- The trend deck: assessment scores / finding counts over time — the programme's heartbeat.
- Finding-to-fix latency by class — the SLA made real.
- AI inventory accuracy: the DSPM AI-app/agent list vs network/endpoint telemetry — the observability coverage, audited against another instrument.
- Sensitive-prompt hotspots: which apps/populations trip detections — the awareness-training targeting data, quantified.