From 1d8ad7aa43ad739c1f3a14ccab71dbc41721c5b0 Mon Sep 17 00:00:00 2001 From: Raja Sekhar Rao Dheekonda Date: Tue, 11 Aug 2026 15:12:10 -0700 Subject: [PATCH] docs(ai-red-teaming): unify capability counts with verified figures Replace stale/inconsistent figures (45 attacks, 500+ transforms, extensive scorer catalog) with verified floors: 70+ attack strategies covering traditional ML and generative AI, 590+ transforms, 140+ scorers, 260 goals. --- .../ai-red-teaming/agents/ai-red-teaming-agent.md | 8 ++++---- capabilities/ai-red-teaming/capability.yaml | 2 +- 2 files changed, 5 insertions(+), 5 deletions(-) diff --git a/capabilities/ai-red-teaming/agents/ai-red-teaming-agent.md b/capabilities/ai-red-teaming/agents/ai-red-teaming-agent.md index 41e032a..8a21a22 100644 --- a/capabilities/ai-red-teaming/agents/ai-red-teaming-agent.md +++ b/capabilities/ai-red-teaming/agents/ai-red-teaming-agent.md @@ -4,8 +4,8 @@ model: anthropic/claude-opus-4-20250514 description: > Unified AI Red Teaming agent for probing security and safety vulnerabilities in LLMs, agentic systems, MCP servers, multi-agent architectures, ML classifiers, and custom AI endpoints. - Orchestrates 45 attack algorithms (41 LLM + 4 adversarial ML samplers), 500+ transforms, - an extensive scorer catalog, and 260 bundled harm goals across OWASP LLM Top 10, + Orchestrates 70+ attack strategies covering traditional ML and generative AI, 590+ transforms, + 140+ scorers, and 260 bundled harm goals across OWASP LLM Top 10, OWASP ASI01-ASI10, and MITRE ATLAS frameworks. --- @@ -205,7 +205,7 @@ When you call `generate_attack`, it: ## Attack Types (common subset) -The capability ships 41 LLM attack algorithms plus 4 adversarial ML samplers; the table below covers the most common picks. Use `"Show me all available attacks"` to enumerate the full set. +The capability ships 70+ attack strategies covering traditional ML and generative AI (generative jailbreaks plus adversarial-ML evasion, extraction, membership inference, and inversion); the table below covers the most common picks. Use `"Show me all available attacks"` to enumerate the full set. | Attack | Best For | Query Budget | |--------|----------|-------------| @@ -241,7 +241,7 @@ The capability ships 41 LLM attack algorithms plus 4 adversarial ML samplers; th ## Transform Catalog -📖 **Complete catalog**: See [transform-catalog.md](./transform-catalog.md) for full reference (500+ transforms across encoding, cipher, persuasion, language, MCP, multi-agent, exfiltration, and more) +📖 **Complete catalog**: See [transform-catalog.md](./transform-catalog.md) for full reference (590+ transforms across encoding, cipher, persuasion, language, MCP, multi-agent, exfiltration, and more) **Common transforms include**: - **Encoding**: `base64`, `hex`, `leetspeak`, `morse`, `unicode_escape` diff --git a/capabilities/ai-red-teaming/capability.yaml b/capabilities/ai-red-teaming/capability.yaml index aaa0514..ba7e960 100644 --- a/capabilities/ai-red-teaming/capability.yaml +++ b/capabilities/ai-red-teaming/capability.yaml @@ -11,7 +11,7 @@ description: > and NIST AI RMF compliance frameworks. Also probes traditional black-box ML classifiers: model extraction (model stealing), membership inference (training-data leakage), and model evasion (adversarial examples) across tabular, image, and text. - 500+ transforms, an extensive scorer catalog, and 260 bundled harm goals across 25 + 590+ transforms, 140+ scorers, and 260 bundled harm goals across 25 sub-categories in safety, security, and agentic tiers. agents: