What security media and vendors’ research teams have published, newest first. A brief is written from what the publisher puts in its feed, says who reports what, and links to the original; where the feed has only a headline, so do we. These are not authorities: read the claims as theirs.
Microsoft Security · Blog ·
Inside the LiteLLM 1.107.0 supply chain compromise: Credential theft through Python startup hooks
LiteLLM 1.107.0 was compromised, stealing credentials.
Microsoft Security reports a malicious LiteLLM 1.107.0 release was published to PyPI, distributing through official container images. The package added files that Python could execute during initialization, targeting authentication material across developer tools and cloud environments. The collector could run without importing LiteLLM, affecting various systems. Microsoft analyzed two execution paths in the affected wheels.
Recommended by the publisher: Microsoft recommends removing the package, addressing the affected Python environment, and credentials accessible to the compromised identity.
Agents are your newest insiders. Security must move at their speed.
Microsoft Security says agents pose insider risk.
Microsoft Security reports agents can access company data and systems, making them a potential insider risk. Agents can make mistakes or be compromised. Microsoft Security is extending Project Perception to secure these agents.
Recommended by the publisher: Microsoft Security recommends surrounding agents with deterministic controls and containment boundaries.
Managing open-weight model risk in enterprise deployments
Microsoft and NVIDIA publish risk management guidance for open-weight models.
Microsoft and NVIDIA say enterprises need to understand model risks in their context. The companies introduce a shared risk taxonomy and approach to managing model risks. They claim weight openness alone does not determine an AI system’s risk.
Recommended by the publisher: Enterprises should apply the guidance to their own use cases, according to Microsoft and NVIDIA.
The hidden network: Inside the phishing infrastructure ecosystem powering global cybercrime
Microsoft uncovered a Chinese-language cybercriminal ecosystem powering global phishing.
Microsoft reports a Chinese-language cybercriminal ecosystem was uncovered, enabling large-scale phishing operations worldwide. Threat actors targeted organizations in healthcare, finance, and IT sectors. Microsoft's research found phishing kits and compromised cloud environments were sold through an underground marketplace.
Recommended by the publisher: Microsoft, in collaboration with law enforcement, disrupted the ecosystem's primary developers and infrastructure.
Post-quantum authentication: Why organizations should start testing certificate ecosystems now
Microsoft says organizations should test post-quantum authentication.
Microsoft says post-quantum authentication affects certificates, trust anchors, and PKI services. Organizations should inventory certificate dependencies and test interoperability. Microsoft’s Post-Quantum Cryptography Transport Layer Security Pilot Program helps with this.
Recommended by the publisher: Microsoft recommends organizations start testing certificate ecosystems now.
Microsoft FORGE Lab discovers Windows vulnerabilities.
Microsoft Security's FORGE Lab reports discovering 140 Windows vulnerabilities. FORGE members also submitted 155 internally validated reports across 23 open-source projects. The lab uses a multi-model agentic scanning harness to organize this work.
CISO perspectives on managing vulnerability risks in the age of AI
Microsoft Security reports AI models change vulnerability management.
Microsoft Security says AI models accelerate vulnerability scanning and patching, affecting CISOs and IT security leaders, who must balance speed and accuracy, and Microsoft infrastructure users, who can deploy Baseline Security Mode to improve security posture.
Recommended by the publisher: CISOs should increase patching resources and rethink patch timing.