ASB
The public AI security evidence index

AI security posture, ranked by evidence.

Compare reviewed open-source AI projects through versioned scores, explicit assurance levels, assessment freshness, and the public evidence behind every result.

Explore the leaderboard How scoring works
Current rankings

Reviewed posture. Interpretation attached.

Every score stays connected to its assurance level, freshness state, assessment date, baseline, and project evidence.

Search and compare all
Reviewed cohort
3 approved projects
Baseline
2026-08-28.static-ai-baseline.v3
Latest assessment
August 28, 2026
Publication rule
Evidence required
01smolagentsAgent FrameworkStatic score81of 100FreshnessFreshAssessed August 28, 2026AssuranceStatic AssessedBaseline: 2026-08-28.static-ai-baseline.v3View evidence
02garakAI Security ToolStatic score78of 100FreshnessFreshAssessed August 28, 2026AssuranceStatic AssessedBaseline: 2026-08-28.static-ai-baseline.v3View evidence
03Inspect AIEvaluation FrameworkStatic score75of 100FreshnessFreshAssessed August 28, 2026AssuranceStatic AssessedBaseline: 2026-08-28.static-ai-baseline.v3View evidence

Static scores describe repository-visible posture under the named baseline. They do not imply runtime validation.

How to read the index

One score. Three explicit evidence levels.

Scoring and freshness
01

Static Assessed

Versioned repository evidence, scoring dimensions, scope, freshness, and limitations.

02

Deep Reviewed

Additional analyst review and editorial validation, clearly separated from the baseline.

03

Runtime Assessed

Selected controls or behaviors tested in a controlled runtime environment.

News & signals

What changed across AI security

Current reporting spans incidents, releases, events, policy, funding, and research. Research is one topic—not the container for every story.

All news
Editorial queue

No reviewed news briefs are published yet.

We will not backfill the launch with invented headlines. The first brief appears after source review, date verification, and editorial sign-off.

See the coverage standard
Case Files

Turn incidents and assessments into reusable lessons

Evidence-led reconstructions of selected failures, controls, and validation work—with the uncertainty left visible.

Open archive
Archive preparing

The first case file is still under evidence review.

Publication waits until the timeline, primary sources, affected control, and limitations can be shown clearly.

  • What happened
  • Why it mattered
  • Evidence and sources
  • Control implications
  • Unknowns and limitations
AI security careers

Put security evidence into practice.

Open-access roles from verified employer application pages—curated for the people building, testing, and governing AI systems.

Browse all jobs
AI Security Engineer

Security Engineer - Detection & Response

LangChain

San Francisco, CA Not specified
View role
AI Security Engineer

Lead Applied AI Engineer

LangChain

New York, NY Not specified
View role
AI Security Engineer

Senior Backend Software Engineer, AI Observability & Evals Platform (LangSmith)

LangChain

Boston, MA Not specified
View role