Transparency

What developers disclose about their models, data and risks.

The average score on Stanford's 2025 Foundation Model Transparency Index fell to 41 of 100, from 58 in 2024. Mandatory disclosure has arrived: EU training-data summaries since August 2025 and California's AB 2013 and SB 53 since January 2026.

Foundation Model Transparency Index, 2025: IBM 95; AI21 Labs 66; Amazon 39; DeepSeek 32; Alibaba 26; xAI 14 FOUNDATION MODEL TRANSPARENCY INDEX, 2025 IBM 95 AI21 Labs 66 Amazon 39 DeepSeek 32 Alibaba 26 xAI 14 SCORE OUT OF 100
Index average, 2024 to 2025
58 to 41
SB 53 penalty per violation
$1M
EU GPAI fine ceiling
€15M or 3%
Registry incidents
2

Evidence

Transparency fell voluntarily and rose by law in the same year

Documented cases: transparency
DateCaseCountryWhat happenedSource
Jul 24, 2025EU training-content summary templateEUProviders list data sources and the top 10% of scraped domainsSource
Dec 2025FMTI 2025United StatesMeta down 29 points, Mistral down 37, OpenAI down 14Source
Mar 4, 2026xAI v. BontaUnited StatesxAI lost its bid to block California's training-data disclosure lawSource
Apr 30, 2026CAC filing updateChina868 generative AI services filed; 17 batches of deep synthesis filings publishedSource

Developers

Meta and all five Chinese developers stayed out of the EU code of practice

Frontier developers
DeveloperModel cardTransparency index, 2025EU code of practice
OpenAI United StatesPartial deployment safety pageabout 37 (derived)Yes
Anthropic United StatesYes system cards46 (derived)Yes
Google DeepMind United StatesYes model card and FSF report41 (derived)Yes
Meta United StatesYes safety report31No
xAI United StatesPartial developer docs only14Partial safety chapter only
Microsoft United StatesUnknownNot assessedYes
Amazon United StatesPartial Nova 2.0 Lite evaluation paper39Yes
Mistral FranceUnknown18Yes
DeepSeek ChinaPartial model card without safety section32No
Alibaba (Qwen) ChinaUnknown26No
Zhipu AI (Z.ai) ChinaPartial launch post onlyNot assessedNo
Moonshot AI ChinaUnknownNot assessedNo
ByteDance (Seed) ChinaPartial Seed 2.0 model cardNot assessedNo

Tag colors: green yes, amber partial, red no, gray not found.

Registry

Two registry incidents bear on transparency

Registry incidents where this discipline was tested in practice. Each links to its full record.

Registry incidents linked to this page
IncidentDateDeveloperRegulator action
Tesla Staff Share Customer Car Camera Footage2019 to 2022TeslaOn 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.
Cruise Robotaxi Drags a PedestrianOct 2023Cruise (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.

Rules

Disclosure is mandatory in the EU, California and China

Rules and frameworks: transparency
RuleJurisdictionWhat it requiresStatus
EU AI Act Article 53 and templateEUPublic summary of training dataFrom Aug 2, 2025; enforcement Aug 2026
California AB 2013CaliforniaTraining-data documentation on 12 topicsIn force Jan 1, 2026
California SB 53CaliforniaTransparency reports and incident reportingIn force Jan 1, 2026
Algorithm registry and generative AI filingChinaFile and publish recommendation, deep synthesis and generative algorithmsOngoing

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 (4 rows); registry incidents (2 linked); developer scorecard (13 developers, static table)
Key sources
EU training-content summary template
FMTI 2025
xAI v. Bonta
CAC filing update
Last verified
October 2026