< ciso
brief />
Tag Banner

All news with #ai supply chain tag

43 articles

House Report Warns US Firms on Chinese Open-Weight AI

⚠️ A new joint congressional report cautions U.S. companies against embedding China-developed open-weight AI models into their technology stacks. Produced by the House Homeland Security Committee and the Select Committee on the CCP, it names Airbnb, DoorDash and Anysphere as examples. The authors compare the trend to Huawei and ZTE, warning that removal could take years and cost taxpayers billions. The report offers 12 policy recommendations to curb the risks.
read more →

Closing the Governance Gap in Open-Weight AI Deployments

🛡️ Open-weight AI is going mainstream, with an estimated 42% of surveyed enterprises running such models in production. Self-hosting offers greater control over data and less provider dependence, but it also shifts security responsibilities onto the organization. Risks include trojanized model files, unclear model provenance, manual updates, and a larger attack surface. The article recommends clear ownership, verified sources, rigorous testing, and isolation before deployment.
read more →

Researchers Hijack Hugging Face Account of Kokoro AI Model

🔐 Researchers took control of the publisher account for Kokoro, a popular text-to-speech model on Hugging Face, by registering an expired domain and resetting the password. They then showed that altering only the model's weights could change its spoken output, with no malicious code for scanners to detect. The case highlights a broader supply-chain trust problem across the open-model ecosystem, as the platform does not verify ongoing account ownership.
read more →

CrowdStrike Details Detection of Backdoored LoRA Models

🛡️ CrowdStrike researchers have outlined a two-stage method for detecting backdoored LoRA models before deployment. Open-weight models are now a core part of the software supply chain, yet organizations have no easy way to verify that a downloaded model was not tampered with. The approach combines static weight analysis with trigger recovery to turn vague suspicion into an actionable finding for analysts.
read more →

Post‑Training Supply Chain Backdoors in Coding Agents

🛡️ Two Microsoft teams demonstrate how post-training supply-chain attacks can implant latent backdoors in coding and terminal agents. Dataset poisoning hides crafted examples in fine-tuning data to teach agents to identify and misuse available command-execution tools, achieving ~78% success in-distribution. Model poisoning assumes control of training and distributed checkpoints, combining SFT and RL to reach 84–95% success on unseen repositories. Both attacks exploit trust in datasets and model weights and highlight the need for data scrutiny, checkpoint provenance, runtime evaluation, and strict tool permissions.
read more →

Prompt Adware: Hidden AI Instructions Raise Enterprise Risk

🛡️ Check Point Research (CPR) uncovered ‘prompt adware’ in a Codex jailbreak bundle called Haiou that persistently modifies an AI agent’s trusted natural-language instructions to inject advertisements for a third-party AI gateway. The finding shows how prompts, Skills and configurations — not executable code — can quietly alter agent behavior, creating visibility and supply-chain challenges for organizations. CPR warns this persistence mechanism could be used for more harmful manipulations, including steering recommendations or exfiltrating sensitive data, and advises firms to monitor agent instructions, vet third-party components and strengthen governance.
read more →

One Evidence Graph Instead of Multiple BOM Programs

🔍 In this analysis, the author argues that enterprises should avoid running separate bill-of-materials (BOM) programs for software, cryptography, AI models, authorization and runtime artifacts. Instead, organizations need a federated evidence graph that links domain-specific records via a common envelope and relationship semantics so incident responders can answer cross-cutting questions about affected products, exposure and remediation. The piece details standards, pilot steps and governance to ensure accuracy, trust and least-disclosure practices.
read more →

How Google Cloud networking supports fluid AI compute

🔍 This article explains how Google Cloud networking accommodates fluid compute for AI workloads, with guidance for choosing GPUs or TPUs and their distinct backend networking needs. It outlines resource obtainment options—such as Flex-start VMs, calendar reservations, future and flex reservations, ComputeClasses, and Spot VMs—and describes four networking configurations: standard networking, accelerated GPU networking (TCPX/TCPXO and RoCEv2), TPU networking, and Cloud Run. The blog highlights deployment blueprints, GKE DRANET automation, and zonal profiles for RDMA, illustrating trade-offs for multi-node training and inference.
read more →

Bidding over Spirit Airlines employee data continues

📰 A dispute over the sale of archived Spirit Airlines employee data persists months after the carrier sought bankruptcy protection. Reportedly including hundreds of millions of emails and chats, OneDrive and SharePoint items, and millions of recorded calls, the dataset has drawn competing bids from AI training firms and at least one major cloud company. Former employees and unions object, citing privacy and unpaid compensation concerns. Stakeholders are considering anonymization as a possible compromise.
read more →

Python package supply-chain risks for AI development

🛡️ On March 24, 2026, widely used Python package LiteLLM was compromised on PyPI, delivering a .pth-based payload that auto-executed on interpreter start. The threat actor group TeamPCP pushed malicious versions that harvested cloud tokens, SSH keys and other secrets, and poisoned packages were available for roughly three hours. The incident is part of a broader rise in malicious open-source packages and highlights unique risks in AI development environments where dependencies can expose models, data and multi-cloud credentials.
read more →

UiPath and Google Cloud: Building a Shared GPU Platform

🚀 UiPath re-architected its infrastructure to support agentic AI and high-scale intelligent document processing by moving from isolated clusters to a shared Google Cloud GPU fleet. The company balances A3 (NVIDIA H100) instances for training with G4 (NVIDIA RTX Pro 6000) instances for inference, using Google Cloud AI Hypercomputer and Dynamic Workload Scheduler to secure predictable capacity. This shared-fleet approach maximizes utilization, reduces costs, and lets engineering teams focus on model performance rather than infrastructure.
read more →

Frontier AI Drives a Surge in OSS Vulnerabilities

🛡️ Unit 42 reports that an autonomous agentic system called NOVA scanned 3,915 open-source projects and found 14,090 confirmed vulnerabilities in two months. The research shows 99.4% of findings were previously unreported and many were high or critical severity, demonstrating how frontier AI accelerates vulnerability discovery and compresses the time between disclosure and exploitation. The report highlights the need for rapid virtual patching, coordinated disclosure, and improved supply-chain and defensive practices.
read more →

OKF v0.2 Adds Frontmatter Trust Signals

📝 OKF v0.2 extends the Open Knowledge Format with optional frontmatter fields that encode provenance, trust, freshness, lifecycle, and attestation signals. The update preserves v0.1's minimalism—new fields are opt-in and backward-compatible—while enabling consumers to filter and assess agent-generated concepts before reading bodies. Reference samples and tooling illustrate attested computations and verification workflows.
read more →

AT&T and Microsoft scale trillion‑token AI workloads

🤝 AT&T partnered with Microsoft Foundry and AMD to build OTel2.0, a telecom-focused AI stack that processes massive token volumes while controlling cost and complexity. Using Foundry Managed Compute and a mix of open models such as Phi-4, OSS-120B, and Gemma-4, AT&T scaled across heterogeneous GPUs (including AMD MI300X) to run experiments, data preparation, and training—processing about 1T tokens and leveraging ~530 GPUs. The effort prioritized rapid deployment, model choice, and economic efficiency.
read more →

Open-source k8s‑aibom for automated AI BOMs

🔍 k8s-aibom is an unprivileged Kubernetes controller that continuously monitors cluster workloads to detect AI runtimes and generate standard CycloneDX 1.6 Machine Learning Bill of Materials (ML-BOMs). It runs as a single Deployment with zero developer friction—no sidecars, no privileged DaemonSets—and exports deterministic BOMs to in-cluster CRs and optional external sinks like Google Cloud Storage. Designed for audit-grade evidence, it supports compliance frameworks and GitOps workflows.
read more →

HalluSquatting: New AI supply-chain attack risks

🛡️ New research describes "HalluSquatting," an attack that exploits AI assistants' habit of inventing resource names. Attackers register those predictable fake names on marketplaces and plant adversarial instructions; when an assistant hallucinate-fetches the same name, it may run the attacker's commands. The technique abuses auto-run modes and agent tools that fetch and execute external code, permitting widespread compromise without traditional malware or exploits.
read more →

How AI Is Rewriting Software Supply Chain Risk

🛡️ Software supply chain security has evolved as AI tools and agents become integral to builds. What used to be a question of third‑party packages and transitive dependencies now includes models, agents, prompts, and autonomous tooling as provenance concerns. Teams must extend lineage to models and pipeline actions, and prioritize findings by actual exploitability to avoid alert overload. The discussion surfaces in a webinar on July 22 covering new research and practical program changes.
read more →

Malicious AI agent skill bypasses security checks

🛡️ A faux AI agent skill called brand-landingpage bypassed static security scanners and reached over 26,000 users via an Instagram ad, highlighting risks as enterprises adopt AI-driven tools. The skill pointed agents to a fake Stitch SDK hosted on a domain controlled by researchers, which initially redirected to the real Google Stitch site to pass review. After distribution, the researchers changed the hosted content to instruct agents to download a script that collected email addresses, demonstrating how mutable external resources let malicious behaviors slip past static reviews. Security vendors and scanners from Cisco, Nvidia, and skills.sh marked the skill safe during testing.
read more →

AI tools surge in underground ransomware marketplaces

🔍 Analysis by Halcyon shows a rapid rise in AI-based tools sold across Telegram channels, dark web forums, and underground markets, with posts increasing from 38 in December 2025 to 1,486 by February 2026. The offerings fall into four groups: weaponized LLMs, AI-enabled identity fraud, AI-augmented malware/infrastructure, and jailbroken or stolen AI services. Ransomware operations are professionalising with tiered services, automation and freemium models, lowering the skill barrier for new actors while law enforcement takedowns and better enterprise defenses remain critical.
read more →

TrapDoor campaign raises developer workstation risk

🛡️ Researchers uncovered the TrapDoor campaign, a cross-registry malicious package operation affecting npm, PyPI, and Crates.io that targets developer workflows and AI coding assistant files. The packages exfiltrated secrets such as AWS credentials, GitHub tokens, SSH keys, browser data, and local dev configs by abusing normal execution points like postinstall scripts, import-time execution, and Rust build scripts. Analysts warn this workflow-focused approach enables persistence and lateral movement into CI/CD and cloud infrastructure, recommending stronger install-time scanning, least-privilege credentials, endpoint hardening, and AI tooling governance.
read more →