Threat Brief — 2026-08-04 — npm Under Siege, AI Weaponises Vuln Data

Executive summary: Two concurrent npm supply-chain attacks — the self-propagating ChainDrop worm (1,300+ packages) and the Keyv-linked credential-stealing worm (353 poisoned versions) — are flooding the JavaScript ecosystem and demand immediate dependency audits. AI is now actively polluting vulnerability infrastructure: 54 hallucinated SQLite CVEs with fake 9.8 ratings have entered NVD, while Unit 42's NOVA system demonstrates industrialised AI-driven zero-day discovery across OSS. Greatness PhaaS has integrated device-code phishing, extending an already surging attack vector into commercial crimeware.

Top items

• ChainDrop npm supply-chain worm compromises 1,300+ packages. A self-propagating malware strain has infected packages with a combined 2 billion monthly downloads on the npm registry. The scale and self-spreading nature make this one of the largest npm poisoning events to date. Immediate dependency pinning and lockfile auditing are advised. (src: BleepingComputer)

• Keyv-linked npm worm poisons 353 package versions, plants Claude Code and VS Code hooks. A credential-stealing worm first seen in keyv@6.0.0 has spread beyond the Keyv and Cacheable namespaces into hundreds of packages across multiple organisations. SafeDep verified 353 poisoned versions across 79 packages. The worm targets developer tooling hooks, making it particularly dangerous for engineering teams using Claude Code or VS Code extensions. (src: The Hacker News)

• AI-hallucinated fake SQLite CVEs with 9.8 ratings pollute NVD. 54 fabricated CVEs claiming critical SQLite vulnerabilities have been submitted to NVD, all rated 9.8 and generated by AI. This degrades trust in vulnerability databases and could trigger wasted patching effort or false urgency. Security teams should cross-reference high-severity SQLite CVEs against upstream advisories before acting. (src: SecurityLab)

• Unit 42 NOVA system discovers 14,000+ unknown vulnerabilities across OSS supply chain. Palo Alto's NOVA demonstrates frontier AI industrialising autonomous zero-day discovery in open-source software. While this accelerates defensive patching, it also signals that the window between discovery and exploitation is shrinking for any actor with comparable AI capability. (src: Unit 42)

• Greatness PhaaS adds device-code phishing to bypass MFA and steal tokens. The commercial Greatness phishing toolkit now supports device-code phishing, abusing OAuth 2.0 Device Authorization to bypass MFA. This extends the device-code phishing surge (first reported 2026-07-31, The Hacker News) into off-the-shelf crimeware, lowering the barrier for attackers. (src: The Hacker News)

• Fake Adobe and Zoom updates install ScreenConnect for persistent remote access. An active multi-wave campaign uses social-engineering lures themed around Adobe/Zoom updates, business document reviews, and system maintenance to stealthily deploy ScreenConnect. Organisations should block unsolicited update prompts and verify software updates only through vendor portals. (src: The Hacker News)

• LLMjacking attacks surge 2.5x as criminals hijack AI compute. CrowdStrike reports a 2.5x increase in attacks by autonomous AI agents, with criminals mining other organisations' neural networks at the victim's expense. Cloud accounts with exposed AI service credentials are the primary attack surface. (src: SecurityLab)

• Russia escalates mass blocking of commercial VPN services. A new wave of restrictions shows that mass VPN disruption no longer requires per-protocol detection or per-operator targeting, suggesting infrastructure-level blocking capability has matured. Operations relying on commercial VPNs in Russia should expect intermittent to total failure. (src: SecurityLab)

• CISA publishes ICS advisory for Acrisure KARR BT and DR-100 vehicle alarm vulnerability. The advisory (ICSA-26-216-01) details a flaw allowing unauthorized vehicle control operations. This formalises the KARR Bluetooth vulnerability first reported on 2026-07-22 by SecurityLab, affecting firmware in an estimated 2M+ vehicles. (src: CISA)

• CISA publishes medical advisory for Thermo Fisher DNA analyzer file-tampering flaw. The advisory (ICSMA-26-216-01) covers a vulnerability allowing modification of .fsa/.hid output files, enabling DNA data tampering and inaccurate test results. This follows initial disclosure on 2026-08-03 by The Hacker News. Affected labs should patch immediately. (src: CISA)

Themes

npm supply chain under coordinated assault: Two independent worms (ChainDrop and Keyv-linked) hitting npm simultaneously is unprecedented. Both are self-propagating and target high-download-count packages. Engineering teams should freeze dependency updates, audit lockfiles against known-clean hashes, and monitor for credential exfiltration from developer environments — especially VS Code and Claude Code hook configurations.

AI as a double-edged sword for vulnerability management: The same day Unit 42 demonstrates AI finding 14,000+ real zero-days, we see AI generating 54 fake CVEs polluting NVD. The net effect: vulnerability pipelines will soon be flooded with both legitimate high-volume AI discoveries and AI-generated noise. Triage automation and upstream-source verification become critical.

Phishing industrialisation continues: Greatness adding device-code phishing to a commercial PhaaS platform mirrors the broader trend of sophisticated techniques (MFA bypass, OAuth abuse) trickling down from bespoke APT tooling to subscription crimeware. Defensive focus should shift to conditional access policies that restrict device-code flow authentication.