diff --git a/.github/workflows/validate.yml b/.github/workflows/validate.yml index 25be197..176838a 100644 --- a/.github/workflows/validate.yml +++ b/.github/workflows/validate.yml @@ -35,6 +35,11 @@ jobs: - name: Run validation run: npm run validate + - name: Check data index is in sync + run: | + npm run generate:data-index + git diff --exit-code lib/data-index.ts || (echo "lib/data-index.ts is stale โ run 'npm run generate:data-index' and commit it" && exit 1) + - name: Type check run: npm run type-check diff --git a/README.md b/README.md index 0be4cf2..cfbb8d0 100644 --- a/README.md +++ b/README.md @@ -7,10 +7,10 @@ **Benchmarks tell you how smart an AI is. TrustVector tells you whether you can trust it in production.** [](https://opensource.org/licenses/MIT) -[](#-current-coverage) -[](/data/models) -[](/data/agents) -[](/data/mcps) +[](#-current-coverage) +[](/data/models) +[](/data/agents) +[](/data/mcps) [](http://makeapullrequest.com) [](https://github.com/guard0-ai/TrustVector) @@ -22,15 +22,16 @@ --- -## ๐จ What our data found (June 2026) +## ๐จ What our data found (July 2026) This isn't a list of logos. Every entry is an evidence-linked evaluation across **security, privacy, performance, transparency, and operations** โ and the findings are uncomfortable: -- โ ๏ธ **21 of 60 models** in the registry are **retired, deprecated, superseded, or were never released** โ including models still hardcoded in thousands of production apps (Grok 3 now *silently redirects* to a different model; Gemini 2.0 Flash was shut down June 1). -- ๐ฉธ **Archived MCP reference servers ship unpatched SQL injection.** The Postgres reference server was still pulling ~21k weekly downloads *after* being archived with a known SQLi โ we score it 51/100 on security so you don't find out the hard way. -- ๐ณ๏ธ **Popularity โ safety.** Context7 (57kโ , the most-starred MCP server on GitHub) scores 86/100 on performance but **59/100 on security** after the "ContextCrush" registry-poisoning vulnerability. Playwright MCP: 88 performance, 60 security. -- ๐ **The agent you let browse the web matters.** General-purpose autonomous agents score as low as **50/100 on privacy** in our registry; sandboxed, permission-gated coding agents score 20+ points higher. -- โณ **The OpenAI Assistants API sunsets August 26, 2026.** If you're on it, your migration window is measured in weeks. It's flagged. +- โ ๏ธ **35+ of 68 models** in the registry are **retired, deprecated, superseded, or were never released** โ including models still hardcoded in thousands of production apps (Grok 3 now *silently redirects* to a different model โ at that model's pricing; Gemini 3 Pro was retired just 4 months after launch; GPT-5 itself has a December 2026 shutdown date). +- ๐ซ **Even the #1 model on the leaderboard can vanish.** Claude Fable 5 โ the highest-scoring model in this registry โ was suspended globally for 19 days in June 2026 under US export controls after a safeguard bypass was found. It's back, but our uptime and jailbreak scores now reflect it. +- ๐ฉธ **Archived MCP reference servers still ship unpatched SQL injection.** The Postgres reference server was still pulling ~21k weekly downloads *after* being archived with a known SQLi โ we score it 51/100 on security so you don't find out the hard way. Meanwhile Langflow, Flowise, and n8n all had critical RCEs actively exploited in 2026 (all patched โ check your version). +- ๐ณ๏ธ **Popularity โ safety.** Context7 (58kโ , the most-starred MCP server on GitHub) scores 86/100 on performance but **59/100 on security** after the "ContextCrush" registry-poisoning vulnerability. Playwright MCP: 88 performance, 60 security. +- ๐ **The agent you let into your life matters.** OpenClaw โ the viral open-source assistant with 300K+ GitHub stars โ scores **37/100 on security** (hundreds of CVEs, 135K+ exposed instances). ByteDance's free Trae IDE scores **24/100 on privacy** (telemetry that survives opt-out, 5-year retention). Sandboxed, permission-gated coding agents score 30-50 points higher on both. +- โณ **The OpenAI Assistants API sunsets August 26, 2026.** That's ~7 weeks out. If you're on it, your migration window is measured in weeks. It's flagged. **Every one of these claims links to a primary source with a date.** That's the whole point. @@ -42,19 +43,21 @@ Overall = mean of 5 dimension scores. Full criteria, evidence URLs, and confiden | Model | Overall | Perf | Security | Privacy | Transparency | Ops | |---|:---:|:---:|:---:|:---:|:---:|:---:| -| **Claude Fable 5** (Anthropic) | **92** | 98 | 92 | 93 | 88 | 91 | +| **Claude Fable 5** (Anthropic) | **92** | 96 | 90 | 93 | 88 | 91 | | **Claude Opus 4.8** (Anthropic) | **92** | 96 | 92 | 93 | 88 | 91 | +| **Claude Sonnet 5** (Anthropic) | **91** | 94 | 91 | 93 | 88 | 90 | | **GPT-5.5** (OpenAI) | **91** | 97 | 89 | 87 | 90 | 94 | | **Gemini 3.1 Pro** (Google) | **91** | 96 | 88 | 88 | 88 | 93 | +| **GPT-5.6 Sol** (OpenAI) | **89** | 94 | 88 | 87 | 87 | 90 | | **Mistral Large 3** (Mistral, open) | **85** | 88 | 83 | 87 | 80 | 86 | -| **Grok 4.3** (xAI) | **83** | 94 | 83 | 76 | 82 | 82 | +| **Grok 4.3** (SpaceXAI) | **83** | 94 | 82 | 76 | 81 | 82 | | **DeepSeek-V4** (open) | **83** | 92 | 83 | 78 | 80 | 83 | | **GLM-5** (Z.ai, open) | **82** | 92 | 80 | 75 | 81 | 83 | | **Kimi K2.6** (Moonshot, open) | **81** | 91 | 79 | 75 | 80 | 82 | Notice the spread: models within 5 points of each other on *capability* differ by **15+ points on privacy and security**. If you're choosing a model for healthcare, legal, or finance, the right-hand columns are the ones that get you fired. -And it's not just models โ the same lens on **coding agents** (Claude Code 80, OpenAI Codex 82, Devin 71, Manus 64) and **MCP servers** (GitHub 82, Playwright 80, Context7 79, archived Postgres 72) exposes exactly where the trust gaps are. +And it's not just models โ the same lens on **agents** (Claude Code 80, OpenAI Codex 82, Devin 71, Claude Cowork 73, ChatGPT Agent 69, OpenClaw 60) and **MCP servers** (GitHub 82, Snowflake 81, Playwright 80, Asana 71, archived Postgres 72) exposes exactly where the trust gaps are.
@@ -130,22 +133,26 @@ Predefined profiles: `balanced` ยท `security_first` ยท `performance_focused` ยท
## ๐ฆ Current Coverage
-**156 evaluations** across 3 categories (last refreshed June 2026 โ yes, including the models that launched *this month*):
+**196 evaluations** across 3 categories (last refreshed July 9, 2026 โ every file re-verified against primary sources, same-day as the Grok 4.5 and GPT-5.6 launches):
{entity.provider}
{entity.description}
@@ -151,10 +144,10 @@ export default async function AgentDetailPage({ params }: { params: Promise<{ id{dimension.data.notes}
} {Object.entries(dimension.data.criteria).map(([key, criterion]) => ( -{criterion.methodology}
-{rating.notes}
@@ -323,20 +310,18 @@ export default async function AgentDetailPage({ params }: { params: Promise<{ id {relatedEntities.length > 0 && ({related.provider}
diff --git a/app/compare/page.tsx b/app/compare/page.tsx index 6a750ce..8c901cf 100644 --- a/app/compare/page.tsx +++ b/app/compare/page.tsx @@ -1,15 +1,14 @@ 'use client'; import { useState, useMemo } from 'react'; -import { getAllEntities } from '@/lib/data'; -import { calculateOverallScore, getScoreColor } from '@/framework/schema/types'; +import { getAllSummaries } from '@/lib/client-data'; import { ScoreBadge } from '@/components/score-badge'; import { GitCompare, X, Plus, Search } from 'lucide-react'; -import type { TrustVectorEntity } from '@/framework/schema/types'; +import type { EntitySummary } from '@/lib/summary-types'; export default function ComparePage() { - const allEntities = getAllEntities(); - const [selectedEntities, setSelectedEntities] = useState
- Compare up to 4 AI models, agents, or MCPs
+ Compare up to 4 AI models, agents, or MCPs
across all trust dimensions.