CISA, the Australian Signals Directorate’s ACSC, and NSA publish joint guidance urging organizations to question and assess third-party vendors on post-quantum cryptography readiness, due to reliance on external providers and the vulnerability of long-lived systems; early, structured vendor engagement and transparency in cryptographic implementations are essential for secure PQC transitions.
Translated title. Original: 「SCS評価制度」★取得を目指すべきか 相談企業の7割が“様子見”する中、明日からできる検討の第一歩 (1/2)
At a Softcreate event, security specialist Toshiya Sakuragi urged companies to assess whether they should prepare for Japan’s planned SCS security rating system. He said its shared criteria could reveal gaps in supply-chain security and reduce the burden of separate security questionnaires.
Researchers reveal a first-of-its-kind supply-chain vulnerability in OpenAI's official skills repository, where trusted skills—contributed or vetted by verified vendors—are used to teach AI agents how to interact with various platforms and services.
Dell emphasizes supply chain security for agentic AI, arguing that agents require new identity and control models, layered security, and end-to-end observability. The company discusses architecture guidance, collaboration with Nvidia, and the need for new tools and skills to secure AI workloads.
Translated title. Original: Dienstleister als Einfallstor: NIS2 bremst den Domino-Effekt
NIS2 seeks to limit domino effects from security breaches in subcontractor networks by enforcing risk-based supply chain management; complex vendor ecosystems can become entry points for cyber threats, including software supply-chain attacks and VPN-based access.
Translated title. Original: Chi sorveglia l'infrastruttura che collega l'azienda all'AI? - Tom's Hardware
LiteLLM faced a supply-chain-style breach: a compromised release containing malware spread via PyPI and Docker, enabling credential theft and remote tool installation. The event underscores the need for strong controls across the AI software stack, including key management, release processes, and operational sovereignty with continuous monitoring.
A compromised LiteLLM release on PyPI hides a credential stealer and backdoor in three files that execute at Python startup; the malicious payload was not present in the official upstream source and the release has been removed, with earlier versions deemed clean. The attack involves injected build data, altered wheel contents, and a background daemon that exfiltrates credentials to a remote server.
Translated title. Original: Quando il malware ha una firma valida: il punto cieco del CRA - Agenda Digitale
A compromised release of the tensorlake npm package carried valid provenance, showing that trusted build records do not guarantee safe code. The article links this risk to malware targeting coding assistants and CI pipelines, and to manufacturers’ supply-chain duties under the Cyber Resilience Act.
Researchers describe CWPdoor, a trojanized backdoor for the Control Web Panel web server, activated by encrypted requests and designed to evade logs; over 340k exposed instances have been identified, with the malware replacing cwpsrv and suppressing traces while enforcing time-bound authentication.
Coordinated vulnerability disclosure drives rapid patching across major vendors, with reputational and market implications depending on the speed and quality of responses; follow-ups include advisories, severity ratings, and potential downstream breach disclosures.