Methodology

Purpose & Scope

Most AI coverage tracks capability releases and funding rounds. This monitor tracks the governance infrastructure — or its absence — that determines whether those capabilities can be deployed legally, safely, and accountably. The target audience is the senior professional who needs to know not just what happened, but what it means for obligations, risk, and strategy in the next 12 months.

What the monitor tracks

• Model capability trajectory — releases, architectural innovations, benchmark threshold events
• Regulatory and legal obligations — EU AI Act implementation, US federal/state law, international standards, active litigation
• Governance gaps — where no binding framework applies and where voluntary commitments are the only constraint
• Compute and infrastructure power — concentration of GPU supply, data centre buildout, energy dependencies
• Personnel flows — movement between frontier labs, safety institutes, and government AI bodies
• Sector deployment — healthcare, legal, finance, defence, media, education, and critical infrastructure
• Information operations — AI-enabled FIMI, synthetic media, and state actor attribution

What the monitor does not do

• It does not provide legal advice or compliance certification
• It does not cover every AI product launch or company announcement — only those with governance, legal, or structural implications
• It does not track AI applications outside the seven covered sectors in M04

Source Hierarchy

All items are sourced according to a three-tier hierarchy. Tier 1 is always preferred; lower tiers are only used when no Tier 1 source exists. The monitor never links to a press article when the primary source is publicly available.

Signal Standard

An item is included if and only if it meets all four criteria. There are no arbitrary item caps — if a module has ten signal-quality items, all ten are included.

Borderline items are assessed against the Asymmetric Signal test: does it contain a non-obvious implication that mainstream press missed or underweighted? If yes, include. Otherwise, omit.

Asymmetric Signal

Every item that warrants it includes an Asymmetric Signal — a non-obvious 12-month implication drawn from technical appendices, regulatory filings, niche research, or academic preprints that the mainstream press missed. This is the monitor's core editorial value: not what happened, but what it means for the trajectory. An asymmetric signal must be specific, actionable for legal/governance/investment professionals, and sourced to T1 or T2.

Friction Analysis

For every legal or regulatory development in Modules 09, 10, and 12, the report includes a Friction Analysis identifying the specific technical capability that the law or standard directly complicates, enables, or outpaces — and the practical implication for compliance. Where International Humanitarian Law is engaged in Module 08, an IHL Friction note is applied.

Confidence Levels

Every item carries an explicit confidence rating. Readers can use this to weight the reliability of an assessment.

Forensic Filters

Six specialist filters are applied every issue. Each is designed to surface a class of development that standard research sweeps routinely miss.

Module Guide

Each weekly issue is structured across 16 modules, numbered 00–15. Every module is covered in every issue. Named sources below are representative of the primary source tier used for each module; they do not constitute the complete source roster.

Cross-Monitor Signals

Each issue includes a Cross-Monitor Signals section identifying where AI governance developments materially overlap with, or are materially affected by, another monitor's domain. A flag is raised when the linkage is structural, not merely topical. Flags follow the same data lifecycle rules as all other persistent entries — a structural linkage is not re-described merely because a week has passed, and closed flags are archived, not deleted. If no material cross-monitor signals exist in a given period, the section states this explicitly; the section is present in every issue without exception.

Data Lifecycle

The monitor builds a cumulative intelligence picture, not a transient news feed. Entries are not deleted because time has passed.

Persistent data

The following remains visible until something material changes:

• Policy positions, legal frameworks, regulatory obligations, and military postures
• Active litigation cases and their procedural status
• Baseline deviations and confirmed risk vector ratings (M07)
• Active monitoring flags — Standards Vacuum, Ciyuan Signal, AISI Pipeline
• EU AI Act 7-layer tracker (M09) and Concentration Index (M14)
• Lab safety policy commitments (RSP, Preparedness Scorecard) and their version history

Transient data

Single announcements, one-off events, dated statements, and tactical incidents may be summarised or rolled into higher-level entries once their implications are captured. They are never silently deleted — they are archived as closed episodes.

Update rules

Version history commitment

Every persistent entry carries a version history recording what changed, when, and why. Past assessments are never silently overwritten. When a persistent state closes — a risk rating drops, a litigation case settles, a standards vacuum is resolved — it is logged as a closed episode with an end date and final assessment.

Limitations

• Geographic scope. Law & Litigation (M09) and European & China Watch (M05) provide depth for EU and US jurisdictions. Other jurisdictions appear in the Country Grid as their regulatory activity crosses inclusion thresholds; they are not comprehensively monitored absent that trigger.
• Capability assessment. Frontier model capability is assessed from public releases and third-party evaluations. Closed research programmes at major labs — and classified military AI programmes — are not directly observable. Chinese lab capability claims are treated as Tier 3 pending independent evaluation.
• Reporting frequency. The monitor publishes weekly. Significant developments occurring mid-week will appear in the following issue, not as breaking updates.
• Analyst judgment. Confidence ratings, risk vector assessments (M07), and asymmetric signals reflect editorial judgment applied to primary sources. They are not algorithmically derived. Readers should treat Speculative-rated items as analytical inference, not established fact.
• Scope boundaries. This monitor does not assess AI safety in a technical sense — it tracks the governance infrastructure around AI, not the alignment properties of specific models. For technical safety evaluation, the relevant primary sources are ARC Evals, METR, and Apollo Research.

Analytical Ecosystem

The AI Governance Monitor is part of the Asymmetric Intelligence network of specialist monitors. The network covers democratic integrity, geopolitical and hybrid threats, conflict escalation, financial crime, environmental risks, global macro, and information operations — alongside AI governance. Cross-monitor signals are issued when findings in one domain have direct structural implications for another. The public methodology pages for all monitors in the network are available at asym-intel.info.

Editor

Peter Howitt · asym-intel.info · Gibraltar

The commons methodology for this monitor is published at asym-intel.info/monitors/ai-governance/methodology/. This commercial edition applies the same research standard with the addition of compliance-actionable filters and the six specialist forensic filters documented above.

Version History