The Risk Index dataset. Every record citable.
The structured companion to the 2026 Risk Index: EU AI Act deadlines, obligation clusters, the seven agent risk dimensions with their actuarial mappings, decided liability cases and the AI insurance market. Every record carries a stable identifier and a traceable source.
01. Enforcement deadline ladder.
Key dates under Regulation (EU) 2024/1689 and the revised Product Liability Directive. The Digital Omnibus, Regulation (EU) 2026/1744, in force since 27 July 2026, moved the Annex III high-risk obligations from 2 August 2026 to 2 December 2027, and Annex I from 2 August 2027 to 2 August 2028. Article 5 prohibitions, Article 4 literacy, Article 50 transparency, the Article 53 and 55 GPAI obligations and the Product Liability Directive were not deferred.
| ID | Date | What applies | Omnibus | Revised date |
|---|---|---|---|---|
| DL-001 | 2024-08-01 | Regulation (EU) 2024/1689 enters into force | No | Not applicable |
| DL-002 | 2025-02-02 | Art. 5 prohibited practices enforceable. Art. 4 AI literacy obligation applies. | No | Not applicable |
| DL-003 | 2025-08-02 | Art. 53 and 55 GPAI obligations apply. AI Office mandate and Art. 99 penalty regime operational. | No | Not applicable |
| DL-004 | 2026-08-02 | Original date for Annex III high-risk obligations (Art. 6, 8 to 15, 26, 27, 43, 72, 73). Deferred by the Digital Omnibus, in force 2026-07-27. | Yes | 2027-12-02 |
| DL-005 | 2026-08-02 | Art. 50 transparency to natural persons. Not deferred by the Omnibus. Chatbot deployers must disclose AI interaction. | No | Not applicable |
| DL-006 | 2026-12-09 | Directive (EU) 2024/2853, the revised Product Liability Directive, transposition deadline. AI software is explicitly a product subject to strict liability. | No | Not applicable |
| DL-007 | 2027-08-02 | Original date for Annex I high-risk obligations (AI in regulated products). Deferred by the Digital Omnibus, in force 2026-07-27. | Yes | 2028-08-02 |
02. Obligation clusters.
Seventeen obligation clusters across Regulation (EU) 2024/1689 and Directive (EU) 2024/2853. Each record names who is bound, the date the obligation applies and the penalty ceiling.
| ID | Obligation | Who it binds | Article | Applies from | Penalty ceiling |
|---|---|---|---|---|---|
| OBL-001 | Prohibited practices | Both | Art. 5 | 2025-02-02 | EUR 35M or 7% |
| OBL-002 | AI literacy | Both | Art. 4 | 2025-02-02 | No direct penalty in isolation |
| OBL-003 | Risk management system | Provider | Art. 9 | 2027-12-02originally 2026-08-02 | EUR 15M or 3% |
| OBL-004 | Technical documentation (Annex IV) | Provider | Art. 11, Annex IV | 2027-12-02originally 2026-08-02 | EUR 15M or 3% |
| OBL-005 | Transparency to deployers | Provider | Art. 13 | 2027-12-02originally 2026-08-02 | EUR 15M or 3% |
| OBL-006 | Human oversight design | Provider | Art. 14 | 2027-12-02originally 2026-08-02 | EUR 15M or 3% |
| OBL-007 | Accuracy, robustness and cybersecurity | Provider | Art. 15 | 2027-12-02originally 2026-08-02 | EUR 15M or 3% |
| OBL-008 | Conformity assessment | Provider | Art. 43 | 2027-12-02originally 2026-08-02 | EUR 15M or 3% |
| OBL-009 | Deployer operational obligations (seven sub-duties including log retention, oversight, incident reporting) | Deployer | Art. 26 | 2027-12-02originally 2026-08-02 | EUR 15M or 3% |
| OBL-010 | Deployer reclassification as provider on substantial modification or rebranding | Deployer | Art. 25 | 2027-12-02originally 2026-08-02 | EUR 15M or 3% |
| OBL-011 | Fundamental rights impact assessment | Deployer (subset) | Art. 27 | 2027-12-02originally 2026-08-02 | EUR 15M or 3% |
| OBL-012 | Transparency to natural persons. Chatbot disclosure, synthetic content marking. Not deferred by the Omnibus. | Both | Art. 50 | 2026-08-02 | EUR 15M or 3% |
| OBL-013 | Post-market monitoring | Provider | Art. 72 | 2027-12-02originally 2026-08-02 | EUR 15M or 3% |
| OBL-014 | Serious incident reporting. 15-day notification to the market surveillance authority. | Both | Art. 73 | 2027-12-02originally 2026-08-02 | EUR 15M or 3% |
| OBL-015 | GPAI baseline obligations (Annex XI technical documentation, Annex XII downstream documentation, copyright policy, training data summary) | GPAI provider | Art. 53 | 2025-08-02 | EUR 15M or 3% |
| OBL-016 | GPAI systemic risk obligations (adversarial testing, risk assessment, AI Office incident reporting) | GPAI provider (systemic) | Art. 55 | 2025-08-02 | EUR 15M or 3% |
| OBL-017 | Product liability. AI software is a product. Strict liability. Reversed burden of proof. | Manufacturers, deployers who modify | Directive (EU) 2024/2853 | 2026-12-09 | Strict liability. No cap. Damage categories include data destruction and psychological harm. |
03. The seven agent risk dimensions.
The canonical dimension set of the Agent Certified methodology v2.0, published 24 April 2026, with names, weights and definitions as in the source. Total weight 100. Each dimension is scored as a whole number from 0 to 10. Scoring formula: the raw score times the weight, summed across all dimensions, divided by 1000 and multiplied by 100.
| ID | Dimension | Weight | Definition |
|---|---|---|---|
| D1 | Trust and Safety | 18, highest | The measurable prevention of unsafe, unauthorised, or harmful actions by the agent in production, and the discipline with which unsafe outputs are detected, contained, and remediated. |
| D2 | Context Integrity | 14 | The quality of the information the agent reasons over, covering provenance, freshness, lineage, and the controls that prevent poisoned, stale, or unauthorised data entering the retrieval pipeline. |
| D3 | Distribution Control | 12 | The controls that determine who can invoke the agent, under what authority, and how its downstream actions are bounded. Where identity, authorisation, and blast radius meet. |
| D4 | Product Maturity | 14 | The degree to which the agent behaves as a production-grade system rather than a prototype, covering reliability, regression discipline, evaluation coverage, and the engineering practices that keep behaviour predictable over time. |
| D5 | Governance | 16, second | The institutional scaffolding around the agent. Evidence that it is known to the board, owned by a named accountable senior role, policed by documented policy, and logged to survive audit. |
| D6 | AI Integration | 12 | How the agent sits inside the organisation's existing systems of record, identity, approval, and escalation. Integration maturity determines whether the agent extends institutional memory or bypasses it. |
| D7 | Autonomy Envelope | 14, critical | The explicit, documented boundary between what the agent may do without human confirmation and what requires a human in the loop. The single clearest determinant of the agent's operational risk profile. |
Mappings
| ID | AI Act | Underwriting line | Best evidence | Gaps closed |
|---|---|---|---|---|
| D1 | Art. 15, 9, 14 | Catastrophic loss line | Dated red team report with kill switch access record | Verification, Governance |
| D2 | Art. 10 | Hidden liability line | End to end data lineage diagram with staleness alerting | Verification, Standards |
| D3 | Art. 26 | Blast radius line | Per tool blast radius assessment with RBAC mapped to the identity provider | Liability chain, European data |
| D4 | Art. 15, 12 | Drift line | Prompt and model version history with regression suite specification | Verification, Standards |
| D5 | Art. 9, 27, 12 | Survivability line | Board minutes referencing the agent plus a risk register extract with a named owner | Governance, Liability chain |
| D6 | Art. 14, 26 | Attribution line | Real audit trail extract showing end to end attribution in a central observability stack | Liability chain, Verification |
| D7 | Art. 14, 26 | PML line (probable maximum loss) | Written action class policy enforced in code, with tested rollback and quarterly sign-off | Verification, Standards; enables PML calculation |
Scoring tiers and insurability
| Tier | Score | Dimension floor | Mark awarded | Insurability read |
|---|---|---|---|---|
| Elite | 75 to 100 | 8 on every dimension | Yes | Preferred risk. PML bounded and documented. |
| Advanced | 55 to 74 | 6 on every dimension | Yes | Standard, accept. Sufficient for most AI liability coverage. |
| Certified | 35 to 54 | 4 on every dimension | Yes | Standard with conditions. Refer on high autonomy deployments. |
| In Progress | 20 to 34 | Not met | No | Referral. Decline autonomous action cover. Typically reaches Certified in 3 to 6 months. |
| Pre Assessment | Below 20 | Not applicable | No | Decline for now. The gap analysis is the roadmap. |
The floor rule. An overall score is an average, and an average can conceal a gap. The per dimension floor rule caps an agent below the tier its weighted total would suggest until every dimension meets the floor. For an underwriter the tier already encodes the worst dimension constraint: no dimension sits in the red at any certified tier.
04. Liability case record.
Decided and pending cases that establish real financial loss from AI outputs. No decided European court judgment establishing AI-specific liability existed as of June 2026. The US and Canadian cases are the precedent and loss patterns underwriters and regulators reason from.
| ID | Case | Citation | Date | Outcome | Loss mechanism | Dimensions |
|---|---|---|---|---|---|---|
| CASE-001 | Moffatt v. Air Canada | 2024 BCCRT 149 | 2024-02-14 | Liability. CAD 650.88 plus interest and fees. Operators cannot disclaim chatbot outputs by framing AI as a separate entity. | Deployer verification gap. No accuracy monitoring after deployment, no mechanism to flag and correct errors. | D1, D5, D7 |
| CASE-002 | Mata v. Avianca, Inc. | No. 22-cv-1461 (S.D.N.Y. 2023) | 2023-06-22 | USD 5,000 sanctions on counsel. Corrective letters to every judge named in fabricated opinions. The gatekeeping duty stays with the professional deployer. | Hallucination. Six non-existent case citations submitted. No verification layer between output and act of reliance. | D1, D2, D6 |
| CASE-003 | Mobley v. Workday, Inc. | 3:23-cv-00770 (N.D. Cal.) | In discovery. Collective certified May 2025. | No merits ruling as of June 2026. Agency theory direct liability claim against the AI vendor survived a motion to dismiss (July 2024). Nationwide ADEA collective certified May 2025. | Algorithmic bias at scale. Potential AI vendor direct liability across all client deployments at once. | D3, D5, D6 |
| CASE-004 | No decided EU AI liability cases | Not applicable | 2026-06-13 | Research found no decided European court judgment establishing AI-specific civil liability as of this date. | EU litigation exists as pending complaints and regulatory investigations. The revised PLD (DL-006) and the EU AI Act provide instruments; case law has yet to form. | Not applicable |
05. AI liability insurance market.
The AI agent liability insurance market as of June 2026. No European-native AI certification body works as an insurance underwriting partner. The US market is well ahead of Europe in standalone AI liability products.
| ID | Player | Role | Status | Coverage or limit | European access |
|---|---|---|---|---|---|
| INS-001 | Artificial Intelligence Underwriting Company (AIUC) | Underwriting company and standards body. Publishes AIUC-1, a security, safety and reliability standard for AI agents. | Active. Emerged from stealth July 2025 with a USD 15M seed led by Nat Friedman, with Emergence, Terrain and an Anthropic co-founder. | AIUC states the standard covers data and privacy, security, safety, reliability, accountability and societal risk. Certificates last 12 months with at least quarterly technical testing. First policy: ElevenLabs, February 2026, after more than 5,000 adversarial simulations. | US-built standard. No European regulatory alignment. |
| INS-002 | Armilla | MGA and Lloyd's coverholder focused on AI liability. | Active. First Lloyd's coverholder dedicated exclusively to AI liability, 2024. Affirmative AI Liability Insurance underwritten by certain underwriters at Lloyd's, including Chaucer. | Up to USD 25M per organisation. Armilla names AI model error liability, AI model output liability and AI agent mistakes among the exposures covered, and states that coverage may not be available in all jurisdictions and is offered only through properly licensed surplus lines brokers. Vanguard AI, launched with Chaucer, combines cyber, technology and AI liability in one coordinated structure. | Via the Lloyd's market through brokers. Not European-native. |
| INS-003 | Munich Re aiSure | AI performance insurance from Munich Re. | Active. Mosaic partnership announced 26 February 2026. | Initial capacity of EUR, USD or CAD 15M via Mosaic. Mosaic describes cover for errors or hallucinations of AI and GenAI models, with loss scenarios including lost revenue, business interruption and legal damages. Claims settle on measurable performance data, without a negligence finding. | For AI developers and vendors worldwide. |
| INS-004 | HSB (Hartford Steam Boiler, Munich Re group) | Standalone AI liability insurance for small and medium-sized businesses. | Active since 18 March 2026. | Standard limits USD 25,000 or USD 50,000, USD 500 deductible, higher limits available. Covers liability for bodily injury and property damage caused by the insured's use of AI, and personal and advertising injury claims arising from AI-generated content. | US product. |
| INS-005 | Coalition | Cyber insurer adding AI endorsements. | Active. Deepfake Response Endorsement added to cyber policies. | Forensic analysis by a deepfake forensics firm, legal takedown work, crisis communications. Available in the United States, the United Kingdom, Canada, Australia, Germany, Denmark, Sweden and France. | Available in Germany and France through cyber policies. |
| INS-006 | Standard-market AI exclusions | Silent AI cover being narrowed in the standard market. | Restricting. ISO endorsement CG 40 47 is reported in trade sources as a standard-form AI exclusion for commercial general liability. This desk has not read the form. | Where an AI exclusion endorsement is attached at renewal, AI-triggered losses fall outside the standard coverage grant. | Creates the gap between mainstream withdrawal and specialist growth. |
| INS-007 | Allianz | European insurer. No standalone AI liability product. | No product as of June 2026. | No product equivalent to AIUC, Armilla or Munich Re aiSure. | European HQ. No European-native AI liability product announced. |
| INS-008 | AXA | European insurer. No standalone AI liability product. | No product as of June 2026. | No standalone AI liability product announced. | European HQ. No European-native AI liability product announced. |
| INS-009 | Zurich | European insurer. No standalone AI liability product. | No product as of June 2026. | No standalone AI liability product announced. | European HQ. No European-native AI liability product announced. |
Market size
| ID | Segment | Estimate | Source attribution | Quality flag |
|---|---|---|---|---|
| MKT-003 | Category formation | Not applicable | Market observation | The category did not exist as a standalone product before 2024. The pace of new entrants from July 2025 to February 2026 confirms structural demand. |
06. National authority status.
National competent authority designation in four primary markets. The EU AI Act is a Regulation and applies directly in every Member State, but Member States had to designate national competent authorities by 2 August 2025. Of 27 Member States, eight met that deadline.
| ID | Member State | Deadline met | Implementing act | Designated authority | Practical status |
|---|---|---|---|---|---|
| NAT-001 | Germany | No (missed August 2025) | KI-MIG, adopted by the Bundestag on 11 June 2026 and in force since 29 July 2026. | Bundesnetzagentur: market surveillance authority, contact point and complaints body. UKIM (AI Market Surveillance Chamber) internal body established. | Implementing law in force since 29 July 2026. |
| NAT-002 | Italy | No. Designated by Law No. 132/2025, in force 10 October 2025, after the 2 August 2025 deadline. | Law No. 132/2025 of 23 September 2025, in force 10 October 2025. First EU Member State to enact a complementing national AI law. | AgID (notifying authority). ACN (market surveillance authority and single point of contact). Banca d'Italia, CONSOB and IVASS keep their market surveillance roles in their sectors. | Most formally advanced. Implementing decrees on civil redress and compliance detail due by 10 October 2026. |
| NAT-003 | Netherlands | No (missed August 2025) | Draft Uitvoeringswet AI-verordening, internet consultation from 20 April to 1 June 2026. Still a draft bill. | Rijksinspectie Digitale Infrastructuur proposed as single point of contact. Autoriteit Persoonsgegevens expected to take a central role for data processing AI. | No single national competent authority formally designated. Decentralised hybrid model proposed. |
| NAT-004 | France | No | DDADUE bill of November 2025: competent authority provisions withdrawn before submission to Parliament. | No formal designation. CNIL proposed for about fifteen AI use cases, ARCOM for media and content, DGCCRF as single point of contact. | Furthest behind of the four reference markets as of June 2026. CNIL is the most operationally active body on AI through GDPR Art. 22 enforcement. |
07. Reference standards and frameworks.
Instruments assessors and underwriters use alongside the EU AI Act as benchmarks. The full numbered reference list is in sources.md.
| ID | Instrument | Published | Type | Role in assessments |
|---|---|---|---|---|
| STD-001 | ISO/IEC 42001:2023 | December 2023 | Standard, not regulation | AI management system requirements. Clause 9 covers performance evaluation; Annex A covers controls. |
| STD-002 | NIST AI RMF 1.0 | January 2023 | Framework, not regulation | Four functions: Govern, Map, Measure, Manage. A structured vocabulary for organisational AI risk governance used by assessors internationally. |
| STD-003 | NIST AI 600-1 | July 2024 | Framework | Twelve named generative AI risk categories. Relevant to D2 Context Integrity for memorisation, synthetic data and training data contamination controls. |
| STD-004 | EIOPA Opinion on AI governance and risk management | 2025-08-06 | Supervisory opinion, not new regulation | Reads Solvency II and the Insurance Distribution Directive for insurance undertakings using AI. Covers data governance, fairness, explainability, human oversight and cybersecurity, and sets the expectations insurers rely on. |
| STD-005 | GPAI Code of Practice | 2025-07-10 | AI Office Code of Practice | Final Code for GPAI model providers in three chapters: transparency, copyright, and safety and security. The Commission and the AI Board confirmed it as an adequate voluntary tool for demonstrating compliance with Art. 53. |