Front Page

Supply Chain

A Trusted Maintainer, a Preinstall Hook, 1.3 Billion Downloads a Month: CHAINDROP Turns Your Own "npm install" Against You

Elastic Security Labs and a coalition of vendors — Microsoft, Palo Alto, Snyk, Wiz, Socket — documented CHAINDROP, the latest strain of the self-propagating Shai-Hulud worm, after attackers compromised the maintainer of the widely used keyv library. The worm inserted a malicious preinstall hook into package.json across more than 400 packages and 2,200 published versions, meaning the infostealer fires the moment a developer runs a routine install or update — no import, no execution of application code required. It harvests npm and GitHub tokens, AWS credentials, Kubernetes secrets, HashiCorp Vault tokens, and Stripe and Slack keys, then uses captured npm tokens to republish trojanized versions of every package the victim can touch.

Command-and-control resolves through an Ethereum smart contract, giving the worm a takedown-resistant address book as it spreads. This is automated supply-chain compromise at a scale that outruns manual triage: a single maintainer account breach cascaded into 1.3 billion monthly downloads of tainted code before defenders could react. Forensic priority: audit CI/CD pipelines and build logs for anomalous npm publish events, rotate every credential type the malware targets — not just npm tokens — and treat any package with a recently modified preinstall script as suspect pending review.

AI Threats

The Documentation Was the Weapon: Attackers Hide Payment Instructions in Fake API Pages to Trick AI Agents Into Wiring Crypto

Zscaler researchers detailed a wave of indirect prompt injection campaigns weaponizing search results and documentation pages against autonomous AI coding agents. Attackers use SEO poisoning to rank fake API documentation for popular developer queries, then hide instructions in the page's DOM using off-screen CSS positioning and JSON-LD metadata — invisible to a human visitor but fully readable by an agent scraping the page for instructions. The payload persuades the agent that a small payment is a required setup step, then routes funds via a Stripe checkout link or a hardcoded Ethereum wallet. Tests showed the technique fooling multiple current-generation models.

Where CHAINDROP automates the exploitation of trust in a package registry, this campaign automates the exploitation of trust in a search result — both aim at the machine doing the reading, not the human. Forensic priority: log and review every URL an AI agent fetches during autonomous tasks, treat unexpected payment prompts or wallet addresses surfaced mid-task as a compromise indicator, and audit agent tool-call transcripts for instructions that originated from fetched content rather than the user.

Threat Bulletin

ACTIVE EXPLOIT
CVE-2026-9198

Langflow OSS — an insecure default lets any network caller mint a SUPERUSER token from /api/v1/auto_login, which is then used to POST Python straight into /api/v1/validate/code for in-process execution via exec(); CVSS 9.8, unauthenticated, no user interaction. Fixed in 1.10.1; added to CISA KEV August 4 after confirmed exploitation. Forensic Note: Review Langflow logs for auto_login calls from unexpected hosts followed by validate/code POSTs, and hunt for dropped payloads on hosts running pre-1.10.1 instances.

ACTIVE EXPLOIT
CVE-2026-18556 / CVE-2026-18577

N-able N-central — an authentication bypass (CVSS 8.2) let attackers gain administrative RMM access and use the platform's Take Control feature to reach managed endpoints; N-able's first patch proved incomplete, and a second bypass (18577) was exploited before the follow-up fix landed. Both added to CISA KEV in early August after confirmed customer compromises. Forensic Note: Audit N-central admin logins and Take Control session logs for anomalous activity, and treat every managed endpoint as potentially touched until confirmed otherwise.

Malware Spotlight

Nova Ransomware — The Rust Encryptor That Blinds Defender First

Analyzed by AttackIQ, Nova is a 64-bit Rust encryptor that opens with a multi-layered defense-evasion routine — reconfiguring Microsoft Defender, killing security services, and terminating any process that looks like a debugger or monitoring tool — before deleting Volume Shadow Copies via vssadmin and WMI. It encrypts with XChaCha20-Poly1305 and RSA-2048, appends the .xgWLckNV extension, drops a README_NOVA.me note, and self-deletes on completion. Forensic Note: Hunt Defender configuration-change events and mass service termination as an early-warning pair, capture volatile memory before self-deletion completes, and recover shadow-copy deletion commands from event logs.