Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 29 additions & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,35 @@ dataset version tracks `package.json` and `CITATION.cff` (kept in sync by

## [Unreleased]

## [0.8.0] - 2026-10-16

### Added
- MITRE ATLAS technique mappings for the whole catalog. 91 of 99 records now
carry `mappings.mitre_atlas`, up from 14, using 58 distinct techniques from
ATLAS 2026.09. The other 8 are left empty on purpose, with a comment: 5 have
`ai_role: disputed`, and 3 describe an agent acting unsafely on its own in a
way no ATLAS technique covers. Where MITRE has published a case study of the
incident, the record includes the AI-native techniques that case study
lists. Four more records now link their ATLAS case study: Morris II
(AML.CS0024), Amazon Q (AML.CS0047), postmark-mcp (AML.CS0053) and
PROMPTSTEAL/LAMEHUG (AML.CS0044), so 10 records link one. Two independent
reviews checked every mapping against the record text before release. The
method is written down in `docs/atlas-mapping.md`.
- `schema/mitre-atlas.json`: the IDs of one ATLAS release (2026.09), generated
from MITRE's release file by `npm run atlas:update -- <file>`.
`npm run validate` now fails on any `mitre_atlas` or
`mitre_atlas_case_studies` ID that is not in that release, and on ATLAS
techniques that ATLAS marks as adapted from ATT&CK (those belong in
`mitre_attack`). Tests cover both rules.

### Changed
- Corrections to the 14 records mapped earlier: AML.T0053 AI Agent Tool
Invocation is removed from GTG-1002 and GTG-2002, because it describes using
a victim's or hijacked agent and both actors ran their own (MITRE's
AML.CS0069 maps GTG-1002 as AML.T0103 Deploy AI Agent). Amazon Q's AML.T0051
is now the more specific AML.T0051.000, as in AML.CS0047. Every other
earlier mapping is kept, with techniques added.

## [0.7.1] - 2026-10-15

### Security
Expand Down
4 changes: 2 additions & 2 deletions CITATION.cff
Original file line number Diff line number Diff line change
Expand Up @@ -10,8 +10,8 @@ authors:
website: "https://mlsecopshub.com"
repository-code: "https://github.com/MLSecOpsHub/agentic-attack-index"
url: "https://mlsecopshub.com"
version: 0.7.1
date-released: 2026-10-15
version: 0.8.0
date-released: 2026-10-16
license: CC-BY-SA-4.0
keywords:
- ai-security
Expand Down
2 changes: 1 addition & 1 deletion CONTRIBUTING.md
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@ npm run archive -- --write # snapshot sources to the Wayback Machine (optional
- The `id` is permanent — it's the citation key. Choose carefully; it is never renamed or recycled.
- One incident per file, one incident per PR.
- YAML style: 2-space indent, ISO 8601 dates, comment any uncertainty.
- `mappings` IDs (MITRE ATLAS/ATT&CK, CVE, AIID) must exist exactly as published — prefer ATLAS for AI-native techniques.
- `mappings` IDs (MITRE ATLAS/ATT&CK, CVE, AIID) must exist exactly as published — prefer ATLAS for AI-native techniques. `npm run validate` checks ATLAS IDs against the release in `schema/mitre-atlas.json` and rejects ATLAS techniques adapted from ATT&CK (record those in `mitre_attack`). See [docs/atlas-mapping.md](docs/atlas-mapping.md) for how to choose techniques.
- Every incident marked `status: confirmed` needs either two sources from **distinct publishers** or a first-party/government source — this is enforced by `npm run validate`.
- Sources rot. Run `npm run archive -- --write` to record a `sources[].archive_url` snapshot; `linkcheck` falls back to it when the live URL later dies. A weekly `archive` workflow also snapshots anything still missing. It never writes to the repo or opens a PR: it opens or updates one `archive`-labelled issue listing the sources still missing a snapshot. A maintainer runs `npm run archive -- --write` locally, reviews the diff and opens the PR.
- Never hand-edit `dist/`; run `npm run build` and commit the result.
Expand Down
2 changes: 1 addition & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,7 @@ Defenders, researchers, journalists, and policymakers who need a grounded, citab

Each record carries a stable `id` (its permanent citation key), a lifecycle-level defensive summary, verification `status` (`confirmed` / `reported` / `test-eval`), sourcing `confidence` (`primary` / `secondary` / `unverified`), and at least one resolvable source.

Records also carry an analytical layer for defenders: an agentic **`autonomy_level`**, a **`guardrail_bypass`** classification (how safety controls were circumvented), an **`ai_role`** skepticism axis (how central AI actually was, independent of status), **MITRE ATLAS / ATT&CK** mappings (ATLAS techniques in `mappings.mitre_atlas`; ATLAS case studies that document the incident in `mappings.mitre_atlas_case_studies`), OWASP references, `related[]` links between records from the same report or campaign, and `mitigations`. `dist/summary.json` rolls these up (by category, severity, status, actor type, autonomy level, AI role, model family, and year) and publishes the dataset's headline honesty figure, **`evidence_split`**: how many records are both `status: confirmed` *and* `ai_role: load-bearing`. Consumers must read that figure rather than recompute it, so every surface quotes the same number.
Records also carry an analytical layer for defenders: an agentic **`autonomy_level`**, a **`guardrail_bypass`** classification (how safety controls were circumvented), an **`ai_role`** skepticism axis (how central AI actually was, independent of status), **MITRE ATLAS / ATT&CK** mappings (ATLAS techniques in `mappings.mitre_atlas`, checked against the ATLAS release in `schema/mitre-atlas.json` and chosen by the method in [docs/atlas-mapping.md](docs/atlas-mapping.md); ATLAS case studies that document the incident in `mappings.mitre_atlas_case_studies`), OWASP references, `related[]` links between records from the same report or campaign, and `mitigations`. `dist/summary.json` rolls these up (by category, severity, status, actor type, autonomy level, AI role, model family, and year) and publishes the dataset's headline honesty figure, **`evidence_split`**: how many records are both `status: confirmed` *and* `ai_role: load-bearing`. Consumers must read that figure rather than recompute it, so every surface quotes the same number.

**Map points (`geo.points[]`, schema 0.3.0).** A record carries map points only where a cited source states a location. Every point has a `role` (origin / target), an explicit **`basis`** saying why it exists (`sponsor-attribution`, `operator-location`, `actor-location`, `infrastructure`, `victim-location`, `stated-location`; see `taxonomy/geo-basis.yml`), the publisher that stated it (`attributed_by`, always one of the record's sources), an ISO `country`, and `illustrative: true` for any centroid. A state sponsor is not an operator location, and `targets.countries` is never turned into points. Consumers must read `geo.points[]`; the pre-0.3.0 `geo.target` / `geo.origin` slots are gone.

Expand Down
2 changes: 1 addition & 1 deletion data/incidents/_TEMPLATE.yml
Original file line number Diff line number Diff line change
Expand Up @@ -146,7 +146,7 @@ lifecycle_phases: []
# mitre_attack for traditional TTPs (e.g. T1566); aiid = AI Incident
# Database incident numbers (integers).
mappings:
mitre_atlas: [] # techniques/tactics/mitigations only (AML.T/TA/M)
mitre_atlas: [] # techniques/tactics/mitigations only (AML.T/TA/M), checked against schema/mitre-atlas.json; method: docs/atlas-mapping.md
mitre_atlas_case_studies: [] # ATLAS case studies of this incident, e.g. AML.CS0059
mitre_attack: []
owasp_llm: [] # OWASP Top 10 for LLM Apps, e.g. LLM01
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -52,7 +52,8 @@ lifecycle_phases:
- initial-access
- impact
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0016.002 # Obtain Capabilities: Generative AI
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand All @@ -74,4 +75,4 @@ sources:
added:
date: 2026-10-10
by: "MLSecOpsHub"
last_updated: 2026-10-10
last_updated: 2026-10-16
5 changes: 3 additions & 2 deletions data/incidents/ai-agents-us-canada-government-websites.yml
Original file line number Diff line number Diff line change
Expand Up @@ -100,7 +100,8 @@ lifecycle_phases:
- initial-access # Attempted only; the SQL-injection probes failed.
- credential-access # Attempted reuse of exposed API keys; no success found.
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0117 # Autonomous Attack-Path Adaptation (Transluce attributes the traffic to agents working public-data tasks; ATLAS includes agents that cross an authorization boundary while pursuing a benign objective)
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand Down Expand Up @@ -143,4 +144,4 @@ sources:
added:
date: 2026-10-15
by: "MLSecOpsHub"
last_updated: 2026-10-15
last_updated: 2026-10-16
6 changes: 4 additions & 2 deletions data/incidents/ai-augmented-fortigate-mass-compromise.yml
Original file line number Diff line number Diff line change
Expand Up @@ -74,7 +74,9 @@ lifecycle_phases:
- credential-access
- privilege-escalation # Domain-level compromise via Active Directory credential extraction.
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0016.002 # Obtain Capabilities: Generative AI
- AML.T0102 # Generate Malicious Commands
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand Down Expand Up @@ -118,4 +120,4 @@ sources:
added:
date: 2026-10-15
by: "MLSecOpsHub"
last_updated: 2026-10-15
last_updated: 2026-10-16
7 changes: 5 additions & 2 deletions data/incidents/aisi-unsanctioned-agent-cyber-testing.yml
Original file line number Diff line number Diff line change
Expand Up @@ -50,7 +50,10 @@ lifecycle_phases:
- initial-access
- deception-social-eng
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0051.001 # LLM Prompt Injection: Indirect (attempted: no model is reported to have ingested the planted content)
- AML.T0117 # Autonomous Attack-Path Adaptation (ATLAS includes agents that cross an authorization boundary while pursuing a benign objective)
- AML.T0118.000 # Autonomous AI Agent Communication: Communication via Shared Artifacts
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand All @@ -72,4 +75,4 @@ sources:
added:
date: 2026-10-10
by: "MLSecOpsHub"
last_updated: 2026-10-10
last_updated: 2026-10-16
2 changes: 1 addition & 1 deletion data/incidents/amazon-kiro-aws-cost-explorer-outage.yml
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,7 @@ lifecycle_phases:
- execution
- impact
mappings:
mitre_atlas: []
mitre_atlas: [] # Not mapped: ai_role is disputed, so an ATLAS technique would assert the AI involvement the sources contest.
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand Down
11 changes: 9 additions & 2 deletions data/incidents/amazon-q-developer-extension-compromise.yml
Original file line number Diff line number Diff line change
Expand Up @@ -38,8 +38,15 @@ lifecycle_phases:
- impact
mappings:
mitre_atlas:
- AML.T0051 # LLM Prompt Injection
- AML.T0010.001 # AI Supply Chain Compromise: AI Software
- AML.T0011.001 # User Execution: Malicious Package
- AML.T0051.000 # LLM Prompt Injection: Direct
- AML.T0065 # LLM Prompt Crafting
- AML.T0081 # Modify AI Agent Configuration
- AML.T0101 # Data Destruction via AI Agent Tool Invocation (attempted: per AWS the injected code failed to execute)
- AML.T0103 # Deploy AI Agent
mitre_atlas_case_studies:
- AML.CS0047 # ATLAS case study "Code to Deploy Destructive AI Agent Discovered in Amazon Q VS Code Extension" (ATLAS v2026.09).
mitre_attack: []
owasp_asi: []
cve:
Expand All @@ -65,4 +72,4 @@ sources:
added:
date: 2026-08-12
by: "MLSecOpsHub"
last_updated: 2026-08-12
last_updated: 2026-10-16
7 changes: 5 additions & 2 deletions data/incidents/anonymouskit-ai-voice-phishing-apple.yml
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,10 @@ lifecycle_phases:
- deception-social-eng
- impact
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0016.002 # Obtain Capabilities: Generative AI
- AML.T0052.000 # Phishing: Spearphishing via Social Engineering LLM
- AML.T0052.001 # Phishing: Deepfake-Assisted Phishing
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand All @@ -66,4 +69,4 @@ sources:
added:
date: 2026-10-10
by: "MLSecOpsHub"
last_updated: 2026-10-10
last_updated: 2026-10-16
Original file line number Diff line number Diff line change
Expand Up @@ -61,7 +61,9 @@ lifecycle_phases:
- execution
- credential-access
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0116 # Autonomous Reconnaissance
- AML.T0117 # Autonomous Attack-Path Adaptation (ATLAS includes agents that cross an authorization boundary while pursuing a benign objective)
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand Down Expand Up @@ -90,4 +92,4 @@ sources:
added:
date: 2026-10-10
by: "MLSecOpsHub"
last_updated: 2026-10-10
last_updated: 2026-10-16
Original file line number Diff line number Diff line change
Expand Up @@ -91,7 +91,9 @@ lifecycle_phases:
- credential-access # Access tokens taken from a site's browser settings file and reused.
- deception-social-eng # Invented homicide tip submitted to police; Anthropic judges the model was producing example content, not trying to mislead.
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0017.001 # Develop Capabilities: Autonomous Exploit Development
- AML.T0117 # Autonomous Attack-Path Adaptation (ATLAS includes agents that cross an authorization boundary while pursuing a benign objective)
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand Down Expand Up @@ -137,4 +139,4 @@ sources:
added:
date: 2026-10-15
by: "MLSecOpsHub"
last_updated: 2026-10-15
last_updated: 2026-10-16
5 changes: 3 additions & 2 deletions data/incidents/artex-south-korea-finance-intrusions.yml
Original file line number Diff line number Diff line change
Expand Up @@ -78,7 +78,8 @@ lifecycle_phases:
- initial-access
- exfiltration
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0016.002 # Obtain Capabilities: Generative AI
mitre_attack:
- T1583.003 # Acquire Infrastructure: Virtual Private Server, per CrowdStrike.
- T1588.007 # Obtain Capabilities: Artificial Intelligence, per CrowdStrike.
Expand Down Expand Up @@ -131,4 +132,4 @@ sources:
added:
date: 2026-10-15
by: "MLSecOpsHub"
last_updated: 2026-10-15
last_updated: 2026-10-16
6 changes: 4 additions & 2 deletions data/incidents/aurora-cursor-agent-ransomware.yml
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,9 @@ lifecycle_phases:
- exfiltration
- impact
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0016.002 # Obtain Capabilities: Generative AI
- AML.T0102 # Generate Malicious Commands
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand All @@ -90,4 +92,4 @@ sources:
added:
date: 2026-10-10
by: "MLSecOpsHub"
last_updated: 2026-10-10
last_updated: 2026-10-16
Original file line number Diff line number Diff line change
Expand Up @@ -93,7 +93,7 @@ lifecycle_phases:
- deception-social-eng # Bank-impersonation vishing lure.
- impact # Bucket defacement, extortion and attempted cryptocurrency drains.
mappings:
mitre_atlas: []
mitre_atlas: [] # Not mapped: ai_role is disputed, so an ATLAS technique would assert the AI involvement the sources contest.
mitre_attack: # As mapped by SOCRadar.
- T1583.001
- T1583.003
Expand Down
6 changes: 5 additions & 1 deletion data/incidents/camoleak-github-copilot-chat.yml
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,10 @@ mappings:
mitre_atlas:
- AML.T0051.001 # LLM Prompt Injection: Indirect
- AML.T0057 # LLM Data Leakage
- AML.T0065 # LLM Prompt Crafting
- AML.T0068 # LLM Prompt Obfuscation
- AML.T0077 # LLM Response Rendering
- AML.T0093 # Prompt Infiltration via Public-Facing Application
mitre_attack: []
owasp_asi: []
# No CVE stated in the sources fetched; some secondary blogs cite CVE-2025-59145
Expand Down Expand Up @@ -68,4 +72,4 @@ sources:
added:
date: 2026-08-12
by: "MLSecOpsHub"
last_updated: 2026-08-12
last_updated: 2026-10-16
9 changes: 7 additions & 2 deletions data/incidents/carbonato-docker-hermes-botnet.yml
Original file line number Diff line number Diff line change
Expand Up @@ -78,7 +78,12 @@ lifecycle_phases:
- credential-access # Agent persona prioritises AI API keys and other credentials.
- impact # Cryptocurrency miner deployed on hijacked hosts.
mappings:
mitre_atlas: []
mitre_atlas:
# The persona file and Telegram tasking show how the implant is meant to work; ThreatDown recovered no agent activity on victims, so command generation and command-and-control use are not mapped.
- AML.T0016.002 # Obtain Capabilities: Generative AI
- AML.T0054 # LLM Jailbreak
- AML.T0081 # Modify AI Agent Configuration
- AML.T0103 # Deploy AI Agent
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand Down Expand Up @@ -107,4 +112,4 @@ sources:
added:
date: 2026-10-15
by: "MLSecOpsHub"
last_updated: 2026-10-15
last_updated: 2026-10-16
7 changes: 5 additions & 2 deletions data/incidents/claude-cowork-file-exfiltration.yml
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,10 @@ lifecycle_phases:
- execution
- exfiltration
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0051.001 # LLM Prompt Injection: Indirect
- AML.T0053 # AI Agent Tool Invocation
- AML.T0086 # Exfiltration via AI Agent Tool Invocation
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand All @@ -63,4 +66,4 @@ sources:
added:
date: 2026-10-10
by: "MLSecOpsHub"
last_updated: 2026-10-10
last_updated: 2026-10-16
7 changes: 5 additions & 2 deletions data/incidents/clawhavoc-clawhub-malicious-skills.yml
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,10 @@ lifecycle_phases:
- credential-access
- exfiltration
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0010.005 # AI Supply Chain Compromise: AI Agent Tool
- AML.T0011.002 # User Execution: Poisoned AI Agent Tool
- AML.T0115.002 # Publish Poisoned AI Artifacts: AI Agent Tools
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand Down Expand Up @@ -79,4 +82,4 @@ sources:
added:
date: 2026-10-10
by: "MLSecOpsHub"
last_updated: 2026-10-10
last_updated: 2026-10-16
12 changes: 10 additions & 2 deletions data/incidents/clinejection-cline-triage-npm-publish.yml
Original file line number Diff line number Diff line change
Expand Up @@ -50,7 +50,15 @@ lifecycle_phases:
- credential-access
- impact
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0010.001 # AI Supply Chain Compromise: AI Software
- AML.T0051.001 # LLM Prompt Injection: Indirect
- AML.T0051.002 # LLM Prompt Injection: Triggered
- AML.T0053 # AI Agent Tool Invocation
- AML.T0065 # LLM Prompt Crafting
- AML.T0093 # Prompt Infiltration via Public-Facing Application
- AML.T0098 # AI Agent Tool Credential Harvesting
- AML.T0119 # Exploit Automated Artifact Processing Pipeline
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand All @@ -76,4 +84,4 @@ sources:
added:
date: 2026-10-10
by: "MLSecOpsHub"
last_updated: 2026-10-10
last_updated: 2026-10-16
6 changes: 4 additions & 2 deletions data/incidents/codewall-mckinsey-lilli-agent-intrusion.yml
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,9 @@ lifecycle_phases:
- initial-access
- credential-access
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0117 # Autonomous Attack-Path Adaptation (ATLAS includes agents that cross an authorization boundary while pursuing a benign objective)
- AML.T0132 # Misconfigured or Publicly Exposed AI Services
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand All @@ -66,4 +68,4 @@ sources:
added:
date: 2026-10-10
by: "MLSecOpsHub"
last_updated: 2026-10-10
last_updated: 2026-10-16
6 changes: 4 additions & 2 deletions data/incidents/coral-sleet-agentic-ai-workflow.yml
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,9 @@ lifecycle_phases:
- resource-dev
- deception-social-eng
mappings:
mitre_atlas: []
mitre_atlas:
- AML.T0016.002 # Obtain Capabilities: Generative AI
- AML.T0054 # LLM Jailbreak
mitre_attack: []
owasp_llm: []
owasp_asi: []
Expand All @@ -71,4 +73,4 @@ sources:
added:
date: 2026-10-10
by: "MLSecOpsHub"
last_updated: 2026-10-10
last_updated: 2026-10-16
Loading
Loading