AI-Infra-Guard v4.6.102026-11-12
[v4.6.10] - 2026-11-12
Added
- Data: Add 19 AI product CVE advisories from NVD (PR #749, c33f4116)
- Compliance: Add a local generative AI safety self-check (PR #748, 970bea9f)
- Data: Add 15 AI product CVE advisories from NVD (PR #745, 29b004de)
- Data: Add 33 AI product CVE advisories from NVD (PR #743, 6ee3cde7)
- Data: Add 35 AI product CVE advisories from NVD (PR #742, 1050abe7)
- Data: Add 33 AI product CVE advisories from NVD (PR #741, 0a480c5c)
Fixed
- Scanners: Handle async local tool execution errors (PR #755, 77d7be03)
- Agent-Scan: Decode batch arrays from XML parameters (PR #754, 14b70330)
- Scanners: Forward optional reasoning effort to the model API (PR #753, a993bd38)
- Scanners: Normalize hex-suffixed tool-call tags (PR #752, 7a1401d3)
- Agent-Scan: Preserve prompt escapes in provider templates (PR #747, c12cf7ac)
- Agent-Scan: Preserve non-object HTTP error responses (PR #746, dcaaad99)
- MCP-Scan: Keep findings that omit optional level/suggestion tags (PR #744, acddd07b)
Changed
- Docs: Document the AI safety compliance check install (29e589c8)
Contributors
2SinceJuly, boyhack, Elon-Jiang, generaltony007, hansenzhf, zhuque
AI-Infra-Guard v4.6.92026-11-05
[v4.6.9] - 2026-11-05
Added
- Agent-Scan: Add CloudGPU provider preset (PR #734, 1c69920f)
Fixed
- Utils: Add timeouts to task-server HTTP clients (PR #740, de5d9f4f)
- MCP-Scan: Accept XML-attribute parameter tags from OpenAI-compatible models (PR #739, 7e6df02a)
- MCP-Scan: Normalize mixed-case MCP SARIF rule IDs (PR #733, 4bb6ac21)
- Agent-Scan: Accept null content deltas and usage-only chunks in SSE replies (PR #731, 851e3f29)
Changed
- Agent-Scan: Remove third-party provider presets (PR #737, 3b12a868)
Contributors
alexcastgonz, boyhack, CloudGPU, Elon-Jiang, Jae Bird
AI-Infra-Guard v4.6.82026-10-29
[v4.6.8] - 2026-10-29
Added
- Scan: Package each scanner for an independent pip install (PR #719, b2135af3)
- Agent-Scan: Add Runcrate provider preset (PR #722, 00a0eed3)
Fixed
- Skill-Scan: Deterministic result for non-standard skill packages (PR #725, ea06b9a5)
- Agent-Scan: Move Hugging Face preset to router.huggingface.co/hf-inference (PR #728, c5224bf0)
- MCP-Scan: Preserve complete MCP tool results (PR #727, 7b62e471)
Changed
- Agent-Scan: Remove legacy MCP client after Agent Provider migration (PR #594, 7e6a48d4)
- Docs: Keep English and Chinese READMEs and document standalone scanner installs (c6526341)
- Docs: Drop README entries that no longer have a section (33768c58)
Contributors
Fulin Zhou, boyhack, Agustin Torres, LuckVd, zhuque, Nirupam Bhowmick
AI-Infra-Guard v4.6.72026-10-22
[v4.6.7] - 2026-10-22
Added
- Agent-Scan: Forward prompt and selected skills to the scanner (PR #704, c11a8527)
- Agent-Scan: Add Cheaper Inference provider preset (PR #720, ee734d11)
- Agent-Scan: Support AG-UI (CopilotKit) protocol streams (PR #710, ba2f8de8)
- Data: Add daily AI CVE advisories and astron-agent fingerprint (PR #717, c6629d4e)
Fixed
- Data: Correct CVE rule ranges and severity vocabulary (PR #716, e13aad6b)
- Fingerprints: Match AutoGen Studio title with or without [Beta] (PR #715, 6fb92e77)
- Fingerprints: Tighten generic matchers and drop non-identifying rules (PR #715, 63a8c95a)
- Yamlcheck: Deduplicate severity labels in isValidSeverity (PR #714, a5c5a72c)
- Models: Avoid panic on unclosed JSON fences with leading text (PR #713, 8e715c38)
- Data: Replace malformed git-hash version rules with real versions (PR #606, 33982126)
Contributors
aiapienthusiast, Cursor Agent, donglige, feiyang, helo060228, stanley, 万家良
AI-Infra-Guard v4.6.62026-10-15
[v4.6.6] - 2026-10-15
Added
- Agent-Scan: Add Atlas Cloud provider preset (PR #706, 6a14a56a)
- Agent-Scan: Add API Route provider preset (PR #697, c752770e)
- Agent-Scan: Add memory/RAG poisoning detection skill (PR #429, a25426cb)
- CLI: Add JSON scan output (PR #682, 3eb7b466)
- Dashboard: Edit extraheaders / extrabody in the model dialog (PR #690, 57fad501)
- Eval: Add BadPhoneAgent bilingual benchmark (PR #698, 0586cb9b)
Fixed
- Agent-Scan: Stream SSE responses and stop at [DONE] instead of waiting for connection close (PR #708, 1e2c69d9)
- Models: Surface terminal ChatStream errors to callers and scan status (PR #702, 1fa464c0)
- Models: Abort instead of retrying after ChatStream emitted partial output (PR #695, 5c3b6a5e)
- Models: Do not panic when a provider returns no choices (PR #692, 6380426f)
- Skill-Scan: Recover missing tool-call args during discovery (PR #648, f74c6d78)
- Runner: Classify severity with the vocabulary the data validator accepts (PR #694, 41bd1321)
- Dashboard: Keep stored extraheaders / extrabody when the update sends the mask back (PR #689, 9f3e846b)
Changed
- Prompt-Security: Drop LLM translation of report case fields (PR #703, 4ea64655)
- CI: Add contributors-readme-action to auto-update contributors list (PR #696, 22b0d385)
- Test (Agent-Scan): Mock the streaming interface in the Atlas Cloud test (3879c190)
Contributors
aigsec, nb213, wjl8636, li872, Devam Shah, helo060228, Truman, Yusheng, boyhack, ymsun2020, drow, DennyHo0917
AI-Infra-Guard v4.6.52026-10-08
[v4.6.5] - 2026-10-08
Added
- Agent-Scan: Support custom transforms for SSE responses (PR #600, f59fbe14)
- MCP-Scan: Support modern/legacy protocol auto-negotiation and record protocol version (PR #579, 890f21dd)
- Agent-Scan: Add OrcaRouter as a first-class provider (PR #619, a7c9b006)
- Models: Send extra headers and body fields to the model API (PR #669, ae6fc0d6)
- DeepTeam: Resume an interrupted red team run (PR #668, a2ac397e)
- Data: Add vulnerability rules [2026-09-25] — 154 CVE rules + 9 fingerprints (PR #679, 1f8ab6f2)
- Data: Add vulnerability rules [2026-10-02] — 117 CVE rules + 5 fingerprints (PR #685, 62a939e1)
Fixed
- MCP-Scan: Rate-limit aware retry for LLM calls (PR #651, 874d05b4)
- Agent-Scan: Tool.inputSchema renamed to input_schema in mcp 2.0 (PR #614, baa3ec51)
- Parser: Keep the pre-release label when normalizing versions (PR #626, 0f5bb1f7)
- Skill-Scan: Keep fallback tool-call parameters with their own call (PR #684, 25e58f10)
- MCP-Scan, Agent-Scan: Bound stalled streaming responses (PR #678, 63c2bd83)
- SARIF: Preserve text remediation in valid results (PR #680, e9e6ed7b)
- Scanners: Preserve findings during context compaction (PR #649, a468ee2c)
- DeepTeam: RailFence crashes with IndexError when rails <= 1 (e39ec6ef)
- Skill file: Fix formatting of description in SKILL.md (PR #580, 8f84377e)
Changed
- CI: Run the Go test suite on push and pull requests (PR #677, ff824825)
- Test: Make agent websocket go test build and pass on main (PR #674, 51d63e24)
Contributors
ichaobuster, BeYoungD, Pawar Priyank, Elwood Zonghao Ying (NY1024), lovejones2914-spec, monem, helo060228, feiyang, Chen Yufeiyang, Jin, Linxiushen (林SO), BevKim, dvd233
AI-Infra-Guard v4.6.42026-10-01
[v4.6.4] - 2026-10-01
Added
- Research: Vendor pikit (https://github.com/NY1024/pikit), a composable prompt-injection research toolkit, under Research/pikit (fe84b853, PR #681)
Fixed
- Skill-Scan: Forward optional reasoning effort to the model API (PR #676, 14beb9b9)
Changed
- Docs: Update Research/pikit README (9f889209)
Contributors
Elwood Zonghao Ying (NY1024), Patrick Erichsen
AI-Infra-Guard v4.6.32026-09-24
[v4.6.3] - 2026-09-24
Added
- DeepTeam: Write finished test cases to disk during a run (#653)
- API Checker: Refill failed fingerprint samples for more reliable model identification
Fixed
- API Checker: Support Claude 5 signatures (PR #667)
- Prompt-Eval: Rate-limit aware backoff, circuit breaker and process-group termination
- Prompt-Eval: Share rate-limit counter between breaker and model, harden Retry-After parsing
- Prompt-Eval: Hoist 'import re' to module top level
Changed
- Research (forgebench)**: Document dataset organization and evaluation-set guidance (PR #662)
- Docs: Add FORGE-Bench and RogueHandoff-20 to Research list (PR #661)
- Docs: Sync FORGE-Bench and RogueHandoff-20 Research entries across 8 language READMEs
- Docs: Add v4.6.2 What's New entry across all 9 README languages (PR #660)
- Test (agent-scan): Pin the detection-skill registration contract (#664)
Contributors
Elwood Zonghao Ying (NY1024), xiangfanwu, KEXNA, Devam Shah, helo060228, aigsec, aig-doc-bot
AI-Infra-Guard v4.6.22026-09-17
[v4.6.2] - 2026-09-17
Added
- Research: Add FORGE-Bench (forge_bench) — deterministic Loss-of-Control benchmark for autonomous agents
- Research: Add clean RogueHandoff-20 benchmark under Research
- Data: Add AIG rules [2026-09-11]
- Skill-Scan: Bump aig-skill-scan to 0.2.2 (#656)
Fixed
- Skill-Scan: Reduce evidence-free false positives (PR #652)
- Skill-Scan: Prevent model tool-call stalls and bound stalled streaming responses
- Skill-Scan: Reconcile verdicts with public findings (PR #652)
- Extract-Vuln: Strip CDATA wrapper and accept common tag aliases (PR #655)
- Data: Flatten cvss map to string format to match upstream schema
- Data: Resolve 12 YAML validation failures
- Data: Remove redundant n8n.yaml fingerprint (upstream has n8n.io.yaml)
Changed
- Research (forgebench)**: Built-in OpenAI-compatible client, bilingual README, and quickstart (PR #658)
- Docs: Document English-only contributor text and canonical input language policy
- Docs: Add v4.6.1 What's New entry across all 9 README languages
- Docs: Revert CATL logo acknowledgement added to all README languages (PR #654/#659)
Contributors
Elwood Zonghao Ying (NY1024), xiangfanwu, aigsec, AIG-Bot, boy-hack, pythoncheng, feiyang, aig-doc-bot
AI-Infra-Guard v4.6.12026-09-10
[v4.6.1] - 2026-09-10
Added
- API Checker: Add Gemini, Gemma, and GLM fingerprints
- API Checker: Update GLM-5.3-Flash fingerprint baseline
- API Checker: Update baseline tests
Fixed
- MCP-Scan: Adapt to MCP Python SDK 2.x API changes (PR #606)
- MCP-Scan: Cast readtimeoutseconds to float for SDK 2.x type safety
- MCP-Scan: Mark empty scan results as possibly-incomplete (PR #623)
- MCP-Scan: Preserve security findings in context compaction prompt
- MCP-Scan: Always set scanNote in result_meta and log in English
- MCP-Scan: Surface the underlying error when MCP connection fails (PR #534)
- Skill-Scan: Stop silently hiding .pyc files and skip-dir payloads
- Task API: Add skill_scan task type
- Parser: Keep pre-release versions below their release in versionCheck (PR #588)
- Data: Correct mislabeled product names for CVE-2026-13236/13237/61428 (PR #642)
- Docker: Use Tencent Debian mirror for agent build
Changed
- Docs: Add CONTRIBUTING, CODEOFCONDUCT, issue/PR templates
- Docs: Update skill-scan README pre_scan capability description
- Docs: Update task API comment clarifying attachments priority in skill_scan
- Docs: Generalize relay consistency check in API Checker docs
Contributors
Elwood Ying, xiangfanwu, aig-doc-bot, valleyxht, kexinoh, zhuque, pucedoteth, glatinone, chaucerj, boy-hack, Ray Winkelman, KEXNA