Intelligence Archives - Phoenix Security

Category: Intelligence

From 40 to 800 Commits: What Breaks in Security When Build Stops Being the Bottleneck

Phoenix’s engineering output jumped from 40 to 800 commits per developer a month. Code review didn’t survive the jump intact — Faros AI’s telemetry on 22,000 developers shows 31.3% of PRs now merge unreviewed. Here’s what broke, and what we rebuilt around generation instead of inspection.

Francesco Cipollone

Exploit Hunt: How Threat-Model-Led AI Hunting Finds Real Zero-Days — and Why It Beats Scanning and Pentesting

Exploit Hunt runs an adversarial three-persona pipeline inside Phoenix Purple’s knowledge graph: attacker, skeptic, exploit developer. Every reported finding ships as a runnable proof of concept, not a severity label. In one verified run, 3 target files produced 9 confirmed exploits — and the skeptic gate correctly killed every disputed finding before it reached proof.

Claire Harwood

Phoenix Security Launches the Exploit Hunt: A Threat-Model-Led AI Red Team That Attacks Your Code and Proves the Exploit

Phoenix Security releases Exploit Hunt at Black Hat USA 2026 — an AI red team that takes targets from a live threat model and reports a finding only after writing and validating a runnable proof-of-concept exploit. Available in Phoenix Purple now.

Francesco Cipollone

Give Me the List of Fixes: Phoenix Purple Launches Graph-Native Remediation for SAST and SCA

Phoenix Purple now runs deterministic SAST and SCA on a knowledge graph, adding reachability, a chainability map, and one-click assessment and remediation. Instead of four disconnected scanner reports, engineers get one ranked fix list with a clear breaking-change verdict on every item, ready to review and ship.

Francesco Cipollone

The Clearinghouse Problem: Coordination Was Never the Bottleneck. Remediation Is.

Gold Eagle solves a coordination gap that was real, but it was never the actual bottleneck. Finding vulnerabilities was never the hard part. The hard part is fixing them everywhere they run, grouped by owner and bundled for remediation velocity. That’s the constraint no clearinghouse touches.

Francesco Cipollone

AsyncAPI Supply Chain Compromise: CI/CD Pwn Request Delivers Miasma RAT Across Four npm Packages

An attacker turned AsyncAPI’s own CI/CD pipeline into a publisher, hiding a PAT theft inside 37 decoy pull requests, then pushing straight to release branches to ship the Miasma RAT to 3M weekly npm downloads under valid SLSA provenance, with zero CVE assigned.

Francesco Cipollone

Close the Tap, Burn the Backlog: The Control Framework Behind Agentic SDLC Security

Francesco Cipollone

easy-day-js / EASY_DAY_JS_MASTRA_2026: Typosquatted Dependency Delivers Cross-Platform RAT to 144 npm Packages

A typosquatted npm dependency called easy-day-js — an exact metadata clone of the legitimate dayjs library — was injected across 144 @mastra packages in an 88-minute automated publishing window, reaching over 1.1 million weekly downloads. The second-stage payload is a cross-platform RAT that installs OS-level persistence on Windows, macOS, and Linux and targets LLM API keys, cloud credentials, and 166 cryptocurrency wallet extensions. No CVE was assigned; every CVE-based scanner was blind during active exploitation.

Sarah Mitchell

The supply chain is under sustained attack. Phoenix Security launches Blue Shield to close the door

Phoenix Security has launched Blue Shield, a behavioural supply chain firewall that blocks malicious packages and AI agent skills at the point of install — across the developer workstation, CI/CD pipeline, and agent session. Built on the Phoenix Blue intelligence backbone, which has tracked 59 campaigns and 657 malicious package versions since June 2024 with zero CVEs assigned during active exploitation, Blue Shield’s free core tier is open today at phxintel.security

Francesco Cipollone

Miasma. The Worm That Lives Inside Your AI Tools

TeamPCP (UNC6780) released Miasma in June 2026: a self-spreading worm that injects itself into the SessionStart hooks of 13 AI coding tools including Claude Code, GitHub Copilot, and Gemini CLI. It forges SLSA provenance signatures to pass npm audit checks, uses GitHub itself as a command-and-control channel, and carries a DEADMAN_SWITCH that wipes developer machines if tokens are revoked before network isolation. Zero CVEs assigned. Every standard scanner returns clean.

Francesco Cipollone

Miasma Worm Reaches Microsoft Azure and PyPI: 73 Repositories Disabled, Hades Wave Drops 37 Malicious Python Wheels

The Miasma worm crossed two new boundaries in 48 hours: GitHub’s automated enforcement disabled 73 Microsoft repositories in 105 seconds after AI coding agent hooks were planted in Azure/durabletask, then 37 malicious PyPI wheels hit 19 packages with .pth startup hooks that steal credentials on every Python invocation. 448 total artifacts tracked. Zero CVEs assigned across the entire campaign.

Marcus Webb

The Acceleration: How Supply Chain Attacks Went Industrial Across npm, PyPI, VS Code, and AI Agent Tooling

Phoenix Security’s Malware Package Intelligence corpus documents 59 supply chain campaigns and 657 malicious package IOCs across npm, PyPI, VS Code, and AI agent tooling from June 2024 through June 2026. The first half of 2026 alone produced 4.5 times the package volume of all 2025 — driven by self-propagating worms, AI assistant config poisoning, and a compiled Rust implant with an eBPF rootkit. Every single campaign: zero CVEs assigned during active exploitation.