Disciplines
10 disciplines, from cybersecurity to ethics
| Discipline | What it covers | Registry incidents | Evidence cases |
|---|---|---|---|
| AI Cybersecurity | Securing AI systems, and AI used to attack and defend. | 5 | 4 |
| Model Safety | Testing frontier models for dangerous capabilities before release. | 3 | 5 |
| Red Teaming | Attacking AI systems on purpose to find failures before others do. | 2 | 5 |
| Alignment | Whether models pursue the goals their developers intend. | 3 | 5 |
| Failure Modes | How AI systems and agents fail: hallucination, miscoordination and misreporting. | 6 | 4 |
| Transparency | What developers disclose about their models, data and risks. | 2 | 4 |
| Explainability | Explaining what an AI system decided and why. | 0 | 4 |
| Accountability | Who answers when an AI agent causes harm. | 6 | 5 |
| Bias and Fairness | Unequal outcomes from AI in hiring, credit, benefits and pricing. | 0 | 5 |
| Ethics | The principles governments, institutions and companies set for AI. | 0 | 5 |
Evidence
Trust in AI is highest in China and lowest in the US among large economies
| Date | Case | Country | What happened | Source |
|---|---|---|---|---|
| Nov 18, 2025 | Edelman flash poll | China, US, UK, Germany, Brazil | Trust in AI at 87% in China, 67% in Brazil, 39% in Germany, 36% in the UK and 32% in the US | Source |
| Apr 2025 | KPMG and University of Melbourne study | 47 countries | 48,000+ respondents: 46% willing to trust AI, 66% use it regularly, 70% want regulation | Source |
| Sep 2025 | Pew US AI survey | United States | 50% more concerned than excited, 10% more excited | Source |
| Jun 13, 2025 | Ipsos AI Monitor | 30 countries | 52% excited and 53% nervous about AI | Source |
| Sep 17, 2026 | Pew, who should regulate AI | 12 countries | China trusted more than the US or EU to regulate AI, 43% against 35% and 34% | Source |
Developers
Every US frontier developer publishes a safety framework; no Chinese developer does
The scorecard records what each developer has published, not how good it is. FLI grades are the Summer 2026 index; the transparency score is Stanford's 2025 Foundation Model Transparency Index. Microsoft, Amazon, Moonshot and ByteDance are not graded by FLI.
| Developer | Flagship model | Safety framework | Third-party testing | FLI grade, Summer 2026 | EU code of practice | Seoul commitments |
|---|---|---|---|---|---|---|
| OpenAI United States | GPT-6 (Oct 2026) | Yes Preparedness Framework v2, Apr 2025 | Yes UK AISI pre-deployment access | C | Yes | Yes |
| Anthropic United States | Claude Opus 5.5 (Sep 2026) | Yes Responsible Scaling Policy v3.4, Jul 2026 | Yes UK AISI cyber evaluation | C+ | Yes | Yes |
| Google DeepMind United States | Gemini 3.1 Pro | Yes Frontier Safety Framework v3.1, Apr 2026 | Yes UK AISI, Apollo Research and others | C | Yes | Yes |
| Meta United States | Muse Spark (Apr 2026) | Yes Superintelligence Scaling Framework v2.1, Oct 2026 | Yes Apollo Research | D+ | No | Yes |
| xAI United States | Grok 4.7 (Sep 2026) | Yes Risk Management Framework, Aug 2025 | Unknown | F | Partial safety chapter only | Yes |
| Microsoft United States | MAI-Thinking-1 (2026) | Yes Frontier Governance Framework, Feb 2025 | Unknown | Not graded | Yes | Yes |
| Amazon United States | Nova 2 series | Yes Frontier Model Safety Framework, Feb 2025 | Partial independent evaluations published | Not graded | Yes | Yes |
| Mistral France | Mistral Large 4 (Oct 2026) | No | Unknown | F | Yes | Yes |
| DeepSeek China | DeepSeek V4 Pro (Aug 2026) | No | No CAISI evaluated after release | F | No | No |
| Alibaba (Qwen) China | Qwen3.7-Max (May 2026) | No | Unknown | D- | No | No |
| Zhipu AI (Z.ai) China | GLM-5.3 (Aug 2026) | No | Unknown | D- | No | Yes |
| Moonshot AI China | Kimi K3 (Jul 2026) | No | No UK AISI assessed after release | Not graded | No | No |
| ByteDance (Seed) China | Doubao Seed 2.0 Pro (Feb 2026) | No | Unknown | Not graded | No | No |
Tag colors: green yes, amber partial, red no, gray not found.
Registry
17 registry incidents test trust in practice
Registry incidents where this discipline was tested in practice. Each links to its full record.
| Incident | Date | Developer | Regulator action |
|---|---|---|---|
| Uber Test Vehicle Kills Pedestrian in Tempe | Mar 2018 | Uber ATG | Arizona suspended Uber's testing on March 26, 2018. Yavapai County declined to charge Uber in March 2019. NTSB adopted its findings on November 19, 2019. The backup operator pleaded guilty to endangerment in July 2023 and received three years of supervised probation. |
| Tesla Staff Share Customer Car Camera Footage | 2019 to 2022 | Tesla | On April 14, 2023, Senators Edward Markey and Richard Blumenthal wrote to Elon Musk demanding Tesla's access policies. An owner filed a proposed class action. No agency penalty was found. |
| Air Canada Chatbot Invents a Refund Rule | Nov 2022 | Not disclosed | On February 14, 2024, the BC Civil Resolution Tribunal found negligent misrepresentation and ordered Air Canada to pay CA$812.02. |
| Cruise Robotaxi Drags a Pedestrian | Oct 2023 | Cruise (General Motors) | California DMV suspended Cruise's driverless permits on October 24, 2023. NHTSA fined Cruise $1.5 million on September 30, 2024, for incomplete crash reports. On November 14, 2024, Cruise admitted filing a false report and agreed to a $500,000 deferred prosecution agreement. |
| Packing Robot Kills a Worker in Goseong | Nov 2023 | Robot installer (not named) | Goseong police investigated, including questioning the on-site safety manager. No ministry action was found. |
| Tesla Recalls Autopilot After NHTSA Investigation | Dec 2023 | Tesla | NHTSA opened PE21-020 in August 2021, upgraded it to EA22-002 in June 2022 and closed it in April 2024 after reviewing 956 crashes, opening RQ24-009 on whether the fix worked. No penalty. |
| EchoLeak Turns Microsoft 365 Copilot Into a Data Leak | Jan 2025 | Microsoft | None reported. |
| Xiaomi SU7 Crashes on Assisted Driving | Mar 2025 | Xiaomi | In April 2025 MIIT barred carmakers from calling driver assistance autonomous driving and required approval of driving-function updates. On September 19, 2025, Xiaomi recalled 116,887 SU7s through SAMR over a highway driver assistance flaw. |
| Claude Opus 4 Blackmails an Engineer in a Test | May 2025 | Anthropic | None. |
| Replit Agent Deletes the SaaStr Database | Jul 2025 | Replit | None reported. |
| Apollo Go Robotaxi Falls Into a Construction Pit | Aug 2025 | Baidu | None found. |
| Stolen Drift Tokens Drain Salesforce Data | Aug 2025 | Salesloft (Drift) | Affected companies filed state breach notices, including a Maine filing covering up to 28,414 people. No regulator action against Salesloft was found. |
| Nx Malware Turns AI Coding Tools Against Their Users | Aug 2025 | Nx; Anthropic, Google and Amazon | None reported. |
| Chinese State Group Runs Espionage Through Claude Code | Sep 2025 | Anthropic | On November 26, 2025, the House Homeland Security Committee asked Anthropic CEO Dario Amodei to testify. |
| Waymo Robotaxis Pass Stopped School Buses | Sep 2025 | Waymo | NHTSA opened preliminary evaluation PE25-013 on October 17, 2025, and wrote to Waymo again on December 3. Waymo counts 12 citations tied to the behavior. |
| Cursor Agent Deletes the PocketOS Database | Apr 2026 | Cursor; model by Anthropic | None reported. |
| OpenAI Test Models Breach Hugging Face | Jul 2026 | OpenAI | Senator Josh Hawley wrote to OpenAI on September 9, 2026. The White House Accord on Super Intelligence, signed September 29, 2026, commits signers to controls that keep models from hacking systems in unintended ways. No enforcement action was found. |
Rules
Commitments are voluntary everywhere; the EU code and California law add teeth
| Rule | Jurisdiction | What it requires | Status |
|---|---|---|---|
| EU GPAI Code of Practice | EU | Transparency, copyright and safety duties for general-purpose models | Published Jul 10, 2025 |
| Frontier AI Safety Commitments (Seoul) | International | Publish safety frameworks with risk thresholds | May 2024; Zhipu the only Chinese signer |
| California SB 53 | California | Frontier frameworks, transparency reports, incident reporting | In force Jan 1, 2026 |
| Interim Measures for Generative AI Services | China | Security assessment and filing before launch | In force since Aug 2023 |
Status colors: green in force, blue voluntary standard or framework, amber adopted and phasing in, gray draft or pending, red withdrawn.
Record
Page Record
- Datasets
- ai_trust_evidence.csv (5 rows); registry incidents (17 linked); developer scorecard (13 developers, static table)
- Key sources
- Edelman flash poll
KPMG and University of Melbourne study
Pew US AI survey
Ipsos AI Monitor
Pew, who should regulate AI - Source notes
- Flagship model names for OpenAI, xAI, Microsoft and Alibaba, and the xAI and Microsoft events, come from secondary sources. Anthropic, Google and OpenAI transparency scores are derived from published changes, not read from Stanford.
- Last verified
- October 2026