The AI landscape doesn't move in one direction — it lurches. Some techniques leap from experiment to table stakes in a single quarter; others stall against regulatory walls, technical ceilings, or organisational inertia that no amount of hype can dislodge. Knowing which is which is the hard part. The State of Play cuts through the noise with a rigorously maintained index of AI techniques across every major business domain — classified by maturity, evidenced by real-world adoption, and updated daily so you always know where you stand relative to the field. Stop guessing. Start knowing.
A daily newsletter distilling the past two weeks of movement in a domain or two — delivered to your inbox while the index updates in the background.
Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail
AI that screens transactions and shipments against export control regulations, dual-use goods lists, and trade restrictions. Includes classification assistance and denied party screening; distinct from sanctions screening which checks entities rather than goods and technology classifications.
AI-powered export control screening has proven operational value at large multinationals but remains bifurcated by organisational scale and integration capacity. The practice applies machine learning to classify goods against dual-use regulations, screen transactions for export restriction violations, and assess end-use risk—distinct from sanctions screening, which checks entities rather than technology classifications. Leading enterprises deploy platforms at scale through SAP GTS, Thomson Reuters ONESOURCE (with AI-powered CoCounsel research across 220+ countries and Gemini-based field extraction achieving 95% accuracy), and Descartes Visual Compliance, automating 95%+ of sanctioned party screening and reducing manual processing 70-80%. Market data shows trade compliance software valued at $2.7B (2025) and projected to reach $9.4B (2036, 12% CAGR), indicating sustained industry investment. Yet adoption fractures sharply at mid-market: cost barriers ($250K-$2M+ implementation), regulatory volatility (May 2025 AI Diffusion Rule rescission, January 2026 H200 policy reversal, May 2026 UK end-use controls), and classification error rates (52% in Japan FY2024 per METI data) create persistent friction. BIS enforcement intensity escalated in May 2026 with GE Aerospace $36M ITAR settlement and concurrent Taiwan raid on Super Micro for chip smuggling, driving compliance investment urgency. Regulatory compliance boundaries expanded with EU AI Act effective August 2026 requiring documented human-in-the-loop oversight and audit trails for trade compliance AI, introducing new deployment requirements. However, government enforcement spending ($122M annually) remains insufficient against documented smuggling diversion ($2.5B in single cases), and U.S. export licensing delays now exceed statutory limits (50%+ of exporters waiting 180+ days for review per CSIS). New regulatory frameworks (UK May 2026 end-use controls, China rare earth controls demonstrating 50% export constraint) accelerate demand for transaction screening, but 63% of mid-market firms report 3-6 week processing delays, indicating capability gaps despite vendor maturity.
The vendor ecosystem consolidates around SAP GTS (95%+ automation in sanctioned party screening, 30-50% cost reduction, 18-24 month ROI), Thomson Reuters ONESOURCE, Descartes Visual Compliance (65,000+ users, 28,000 annual man-hours saved), and Oracle GTM, with mature SaaS deployments accelerating through cloud integration (AWS Marketplace availability). Emerging vendors target agentic classification: GingerControl (2026 GA, founded by Flexport/SAP/Customs/Oracle leaders) offers AI-native HTS classification with transparent reasoning chains; KYG Trade (2026 AI-native GTM² platform) features agentic JCAP (Jurisdiction and Classification) with real-time regulatory monitoring and named customer metric (50% classification time reduction for $18B OEM). Certivo products automate rare earth origin tracking and ITAR compliance documentation, signaling deployment across regulated commodity domains. AI application broadens beyond classification to anomaly detection and entity relationship mapping: Kharon ClearView reduced Military End-User screening time by 90% through visualized entity risk networks. Enterprise adoption continues: Resonac's ONESOURCE rollout across 220+ jurisdictions, Thomson Reuters CoCounsel reaching one million professionals (including trade compliance specialists), and consulting adoption (KPMG SAP GTS implementations) all signal production-grade deployment. Gartner notes trade compliance solutions remain "typically siloed and tactically deployed" rather than strategically integrated, indicating continued maturation pathway.
Enforcement environment escalated significantly in May 2026. GE Aerospace faced $36M ITAR settlement with 36-month Consent Agreement for 116 violations including manual overrides of automated export compliance systems, demonstrating high enforcement visibility into operational gaps. Concurrent Taiwan raid on Super Micro (12 locations, 3 individuals targeted) for routing Nvidia AI chips to sanctioned jurisdictions signals downstream partner liability expansion. BIS and OFAC enforcement action remains acute: Applied Materials $252.5M settlement (February 2026) for 56 unauthorized reexport charges through Korea routing; Coastal PVA Technology $1.7M penalty for EAR99 semiconductor brush exports; Teledyne FLIR $1M for specification changes. OFAC penalties surged to $265M in 2025 (up from $49M in 2024). These penalties drive compliance investment: 43% of trade professionals increased hiring, 38% boosted technology spending, 34% expanded training (Thomson Reuters 2026 survey). However, a critical enforcement-policy gap persists: federal enforcement spending ($122M annually) trails smuggling diversion ($2.5B in a single case). Justice Department prosecutions of smuggling networks exposed serial-number spoofing and sophisticated transshipment routing through South Taiwan subsidiaries and Southeast Asia data centers. Compliance-side barriers remain acute despite enforcement intensity: 50%+ of U.S. tech exporters wait 180+ days for export licenses (CSIS May 2026), well beyond 90-day statutory limit; 63% of mid-market firms report 3-6 week export compliance validation delays; practitioners document widespread single-regime screening (OFAC OR BIS but not both), creating blind spots. Regulatory frameworks expanded: UK activated May 13, 2026 "end-use controls" framework requiring licenses for goods after exporters are "informed" of diversion risk, with £1M or 50% breach-value penalties. Japan METI data shows 52% classification error rate (FY2024), with deployed TRAFEED AI solution achieving ~70% determination time reduction. China rare earth controls demonstrate real market constraint: yttrium exports to US fell 95% (333 to 17 metric tons), aerospace manufacturers report shortages. The enforcement-capacity gap directly drives demand for agentic classification and advanced transaction screening positioned as force-multipliers for under-resourced compliance teams. BIS criminal prosecution remains active (26 DTCF cases, up to 20-year liability). Regulatory uncertainty persists with no replacement framework for May 2025 AI Diffusion Rule rescission, though January 2026 case-by-case H200 review and UK May 2026 end-use controls signal pragmatic regulatory recalibration amid enforcement limitations.
Government-scale deployment evidence emerged in July 2026: Shanghai Customs' smart clearance pilot reduced CNC export processing from 48 hours to 2.3 hours via AI document recognition and pre-classification; TIMEWELL's TRAFEED AI system achieved 95.5% classification accuracy with 99.7% time reduction (2–3 hours to 5 seconds); Ukraine's State Customs Service integrated LLM-based classification into official customs portals. However, ACL 2026 peer-reviewed research (HSCodeComp benchmark) revealed a critical capability limit: AI agents achieve only 49.4% accuracy on 10-digit HS code classification versus 95% for domain experts—a 45-point gap persisting despite test-time scaling and multi-voting. This structural limitation, rooted in cascading errors across hierarchical rule systems, underscores that while AI excels at high-volume screening and first-pass classification, expert validation remains mandatory for edge cases and nuanced regulatory interpretation. Simultaneously, enforcement infrastructure faces operational constraints: BIS licensing delays increased to 180-300 days (versus 38-day historical baseline) and the Entity List remained frozen 8 months, limiting the deterrent effect of regulatory machinery even as violations accumulate.
A deeper challenge undermines control sustainability: market pressures are eroding policy consistency. R&D funding has shifted from public (65% during Cold War) to private sector (75% today), and major technology firms derive 20-50% of sales from China-dependent customers, creating economic incentives that trigger policy oscillation and seesawing enforcement intensity. This structural tension produces control regimes that simultaneously fail to contain technology diffusion while destabilizing U.S. innovation investment. Additionally, the June 12, 2026 Anthropic export control order exposed institutional design gaps in export-control governance: the process lacked validation, procedural fairness, and proportionality safeguards, with remediation requiring external litigation and legislative proposals for structured institutional review with written determinations and judicial oversight. These tensions—strong vendor ecosystem maturity coupled with documented AI limitations, enforcement delays, market-pressure-induced policy volatility, and governance process failures—define the current deployment landscape and constrain mid-market adoption despite technical platform maturity.
— Major vendor Q1 FY2027 earnings: record revenue $193.6M (+15% YoY), 46% EBITDA margin, with explicit AI agent layer expansion across sanctioned-party screening, routing, dispatch, and research agents.
— Zhong Lun analysis of China's expanding export-control framework: Provisions of State Council (July 1, 2026) embed technology-transfer screening into M&A and outbound investment, paralleling US/EU regulatory sophistication.
— Market analysis distinguishing agentic (end-to-end automation of classification, duty, screening, filing) from assistive (flag for human review) platforms; shows trade compliance software maturity spectrum with deployment model divergence.
— TIMEWELL/TRAFEED deployment metrics: 95.5% classification accuracy (Okayama University PoC with consensus AI + rule engine), 2–3 hours per case reduced to 5 seconds (99.7% time reduction), addressing Japan's 52% security-trade violation error rate.
— Academic synthesis of fragmented global AI export-control regime (US BIS, Dutch DUV, Japanese METI, Chinese countermeasures); identifies June 12-13, 2026 Anthropic directive as first nominative export-control order on commercial AI model already in market.
— CASRAI analysis of 2026 BIS enforcement patterns including Applied Materials $252.5M settlement (statutory maximum for semiconductor reexports to China), identifying avoidable violations: unlicensed reexports, missed deemed-export screening, shipments without authorization.
— Critical Lawfare analysis: export-control regime fails due to market pressures and economic dependencies; R&D funding shift (Cold War: gov 65%, 2024: businesses 75%), Qualcomm 50% China sales, causing oscillating policy undermining control effectiveness.
— Strategic analysis of June 2026 Anthropic export control order: process lacked validation, procedural fairness, proportionality safeguards; calls for statutory AI security review agency with structured institutional review, written determinations, and judicial review.