TL;DR: AI adoption is being constrained less by model quality than by telemetry quality, with Axoflow arguing that schema drift and normalization ownership determine whether data is usable for detection and agentic AI. The operational lesson is that pipeline control, not just visibility, now sets the effective security ceiling.
At a glance
What this is: Axoflow says telemetry normalisation, not model quality, now determines whether security data is usable for detection and AI.
Why it matters: For SOC, IAM and AI security teams, the issue is whether changing log formats and brittle pipelines silently undermine detection fidelity, governance and agentic AI inputs.
By the numbers:
- Only 23% have the infrastructure to back that up.
Context
Telemetry normalisation is the control layer that makes security data usable after collection. When vendors change log formats, rename fields or truncate output, downstream detections and AI consumers inherit the breakage unless the pipeline owns normalisation.
The governance gap is ownership. Many environments can route data, but far fewer can guarantee that every source arrives classified, enriched and structurally consistent without continual human repair. That matters for SOC operations today and for agentic AI systems that will be only as reliable as the telemetry they consume.
Key questions
Q: What breaks when telemetry is not normalised before analysis?
A: Cross-source correlation becomes unreliable, identity and network events are harder to match, and AI tools have to learn around inconsistent labels instead of analysing behaviour. The result is slower investigations, weaker automation, and more manual interpretation by analysts.
Q: When should organisations prioritise normalization over more telemetry volume?
A: They should prioritise normalisation when collection is already broad but downstream use cases still fail because fields are inconsistent or missing. More volume does not fix structural ambiguity. If the security team cannot trust the shape of the data, adding more of it only increases cost and noise.
Q: How do security teams know whether telemetry normalization is working?
A: It is working when the same investigative question can be answered consistently from Windows, cloud, and identity sources without custom parsing for each consumer. Analysts should be able to trace a normalized event back to its original source, preserve key fields such as timestamps and origin, and reuse the same model across detections and investigations.
Q: Why do agentic AI systems depend on normalized telemetry?
A: Agentic AI needs stable, enriched data because it cannot make trustworthy decisions from inconsistent event structures. If the underlying telemetry is drift-prone, the model or agent inherits uncertainty at the point of action. That makes data governance a prerequisite for AI governance, not a separate concern.
Technical breakdown
Why schema drift breaks detections silently
Schema drift is the steady change in event structure across log sources, vendors and collection paths. A field can move, rename, split or disappear without any obvious pipeline failure. Detection content that depends on a specific field path may still compile, but it no longer matches the incoming record. That creates a quiet failure mode: the telemetry arrives, the SIEM ingests it, and the rule simply stops firing. In practice, the issue is not collection volume but structural consistency across hundreds of source variants.
Practical implication: Treat schema drift as a detection integrity problem, not a formatting nuisance.
Why normalized telemetry is the real AI data floor
AI systems do not repair broken telemetry semantics. If data arrives unclassified, inconsistent or incomplete, an AI layer can only infer around the gaps, which weakens detection quality and response confidence. Normalisation is the process of turning heterogeneous logs into a stable structure with predictable fields, labels and enrichment. For agentic AI, that floor matters even more because the system may take action on the data it receives. If the underlying event model is unstable, the AI stack inherits ambiguity instead of clarity.
Practical implication: Build AI pipelines on data contracts that guarantee stable structure before any model or agent consumes the telemetry.
Why pipeline ownership matters more than visibility
Visibility tools tell teams what is moving through the pipeline, but ownership determines who fixes it when source formats change. If every vendor update becomes an internal parser ticket, the organisation has not removed complexity. It has only shifted where the work happens. The stronger model is to make normalization a managed control rather than a customer-maintained rule set. That changes the operating model from constant repair to predictable governance, which is what both SOC teams and AI-enabled workflows need to scale.
Practical implication: Assign clear ownership for normalization logic so source changes do not become recurring detection outages.
Threat narrative
Attacker objective: The practical attacker gain is reduced detection fidelity, which creates blind spots that can delay investigation and response.
- Entry occurs when vendor log formats change or fields are truncated, and the pipeline accepts the altered structure without a visible failure.
- Credentialed access is not the issue here, but the equivalent control failure is ungoverned schema drift that bypasses the assumptions in detection content.
- Impact appears when rules stop matching silently, leaving investigations to discover gaps only after a missed alert or delayed response.
Breaches seen in the wild
- reviewdog Action compromise 2025: A stolen maintainer token poisoned reviewdog/action-setup, leaking CI secrets including the tj-actions bot token used in the next attack.
- ArtiPACKED 2024: Unit 42 found public GitHub Actions artifacts in major open source projects leaking live GITHUB_TOKENs, runtime tokens and cloud tokens.
Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.
NHI Mgmt Group analysis
Telemetry normalisation is now a governance control, not a data engineering convenience. Once detection content and AI systems depend on structured fields, the organisation is no longer managing a logging problem, it is managing control integrity. A pipeline that silently tolerates drift can invalidate both human detections and machine consumption without any explicit outage. Practitioners should treat normalisation ownership as part of security architecture, not a support function.
The new failure mode is operational invisibility, not missing telemetry. Security teams can collect enormous volumes of data and still lose efficacy if the structure changes underneath their rules. That is why the article’s central point matters: the issue is who absorbs the cost of format change. In practice, the control gap is not observability, it is unowned transformation logic that turns every upstream change into a downstream security debt.
Telemetry quality is the prerequisite for agentic AI trust. Agentic AI systems do not create value from raw intake alone, they depend on data that is already classified, normalized and enriched. If the input contract is unstable, the AI layer amplifies ambiguity rather than reducing it. This directly intersects with identity and governance because machine decisions are only as trustworthy as the data provenance and structure they consume. Practitioners should make telemetry stability a gating control for AI adoption.
Schema drift should be treated as a named security concept: detection integrity decay. The phrase captures what happens when small source changes steadily erode the reliability of rule-based detection and AI-ready telemetry. It is not a single incident, but a cumulative governance failure that widens blind spots over time. The right response is to measure whether the pipeline preserves meaning across source changes, because that is the point where the control either holds or quietly fails.
From our research library:
- 67% of organisations still rely heavily on static credentials despite the risks they pose to agentic AI deployments, according to the 2026 Infrastructure Identity Survey.
- 53% of security leaders expect AI to run major portions of their infrastructure autonomously within the next three years, according to the 2026 Infrastructure Identity Survey.
- Read next: Agentic AI Identity Maturity Model
What this signals
Detection integrity now depends on who owns the parser. If source systems can change field structure without forcing a governed update, the security programme is effectively accepting silent control decay. That is why telemetry normalisation belongs in the security architecture conversation, not just the observability stack.
Agentic AI raises the cost of structural inconsistency. Models and agents can only act confidently when the telemetry layer preserves meaning across source changes. According to the 2026 Infrastructure Identity Survey, 67% of organisations still rely heavily on static credentials despite the risks they pose to agentic AI deployments, which underscores how immature many foundations still are.
Telemetry normalisation is a prerequisite for machine trust. When the pipeline owns classification, enrichment and schema stability, SOC teams can shift effort from repair to detection engineering. That is the operating model change practitioners should expect as AI-enabled security workflows mature.
For practitioners
- Define ownership for normalization logic Assign explicit operational ownership for parsing, field mapping and enrichment so source changes do not create ambiguous repair work.
- Track schema drift as a security event Alert when a source changes field names, truncation patterns or event structure in a way that could break detections.
- Validate detection content against normalized fields Test rules against the post-normalization event model, not only against raw source samples, so silent mismatch is caught before production.
- Separate telemetry storage from detection inputs Keep raw logs for retention and forensics, but ensure downstream detections and AI agents consume only stabilized, enriched data.
- Gate agentic AI on telemetry quality Require stable classification, enrichment and structural consistency before allowing agents to act on operational security data.
Key takeaways
- Telemetry quality, not model capability, is the binding constraint when security data must support both detection and AI consumption.
- Schema drift creates silent control failure because detections can keep running while the fields they expect have already changed.
- The practical answer is to treat normalization ownership as a security control and validate AI inputs only after telemetry is structurally stable.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
MITRE ATT&CK and OWASP Agentic AI Top 10 address the attack and risk surface, while NIST CSF 2.0, CIS Controls v8 and NIST AI RMF set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| MITRE ATT&CK | TA0007 — Discovery | Telemetry drift hides failed detections and delays discovery of security gaps. |
| Recommendation — Map silent telemetry failures to discovery gaps and monitor for sources whose structure no longer matches rules. | ||
| NIST CSF 2.0 | PR.DS-01 — Data-at-rest is protected | Normalized telemetry is a data handling control that preserves integrity for downstream security use. |
| Recommendation — Protect telemetry integrity with governed transformation controls before it reaches detection and AI systems. | ||
| CIS Controls v8 | CIS-8 — Audit Log Management | The article is about making logs reliable enough for security monitoring and response. |
| Recommendation — Centralize audit log management and validate that parsing changes do not break monitoring coverage. | ||
| NIST AI RMF | GOVERN — AI Governance and Accountability | The article ties AI readiness to data governance and accountable ownership of the telemetry layer. |
| Recommendation — Assign accountable ownership for AI input data quality before deploying agentic systems. | ||
| OWASP Agentic AI Top 10 | ASI03 — Identity & Privilege Abuse | Agentic systems can only act safely when their inputs are stable and trustworthy. |
| Recommendation — Treat unstable telemetry as a privilege-risk amplifier and gate agent actions on verified inputs. | ||
Key terms
- Schema Drift: Schema drift is the mismatch between the attributes an IdP sends and the fields an application can store or interpret. It often appears as missing custom fields, inconsistent group data, or varying attribute names, and it undermines the reliability of lifecycle automation even when the core protocol works.
- Telemetry Normalization: Telemetry normalization is the process of turning data from different security tools into a consistent format that can support one policy decision. It is essential when identity, endpoint, and asset systems all feed the same control plane, because conflicting data can otherwise create gaps or overblocking.
- Detection Integrity: The degree to which an organisation’s alerting and monitoring system continues to surface real threat activity after tuning, suppression, and automation changes. It is not the same as low alert volume. A system can look quieter while becoming materially less trustworthy for security decision-making.
- Agentic AI Readiness: Agentic AI readiness is the state of having enough identity visibility, access governance, and auditability to let autonomous systems operate without relying on unknown shortcuts. It is not a model-quality measure. It is a control-state measure that asks whether the environment can support machine-led action safely.
Deepen your knowledge
The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, agentic AI identity and workload identity for practitioners shaping modern security programmes. It helps teams align identity controls with the operational realities of machine and AI-driven systems.
Published by the NHIMG editorial team on August 18, 2026.
Updated on October 12, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org