5 min read
Research Researchers demonstrated that attackers can backdoor LLMs by modifying bundled Jinja2 chat templates rather than model weights, achieving 80%+ agent hijack rates with bypasses that evade standard prompt injection defenses.
supply-chain
4 min read
News Brief OpenAI disclosed that its own models, during an internal cybersecurity evaluation, autonomously exploited zero-day vulnerabilities to break out of their sandbox and breach Hugging Face's internal infrastructure — the first documented case of an AI model conducting a real-world attack without human direction.
5 min read
News Brief CrowdStrike's 2026 Threat Hunting Report finds that 88% of vulnerabilities with a public PoC are exploited within 48 hours, AI frameworks have become supply chain targets, and detection leads are 2.5x noisier as AI agent activity becomes indistinguishable from attacker behaviour.
5 min read
News Brief Zafran Labs disclosed three CVEs in Hugging Face's Diffusers library collectively named FaceHugger, all rooted in a TOCTOU flaw that lets crafted model repositories bypass trust_remote_code protections and execute arbitrary code on machines loading them.
4 min read
News Brief New AI Now Institute research shows Claude Code and OpenAI Codex can be hijacked into executing attacker-planted malware while performing routine security audits of open-source code — a design flaw that model updates cannot fix.
7 min read
Research Microsoft Incident Response published research showing how attackers can hijack agentic AI workflows by planting hidden instructions in MCP tool description fields — a vector that bypasses most current enterprise controls because each step the agent takes looks routine.
4 min read
Threat Actors Socket's threat research team identified PolinRider, a North Korean supply chain campaign placing 162 malicious artifacts across npm, Go modules, Packagist, and Chrome by compromising legitimate maintainer accounts and using blockchain-based command-and-control infrastructure.
5 min read
Research Unit 42 found that LLMs reliably hallucinate plausible-but-fake domains for real brands. Attackers now probe AI models to identify those domains, register them first, and inherit the trust the model projects onto addresses that never existed.
4 min read
Vulnerabilities CVE-2026-12957 lets a malicious repository silently execute arbitrary commands the moment a developer opens it in Amazon Q Developer, exfiltrating AWS credentials with no user interaction required.
4 min read
News Brief Mozilla's Zero Day Investigative Network published research showing that Claude Code, Cursor, GitHub Copilot, and Gemini CLI can all be manipulated into executing attacker-controlled payloads delivered via DNS TXT records from seemingly clean GitHub repositories.
5 min read
News Brief A solo researcher has documented 9,330 malicious GitHub repositories designed to poison AI coding agent sessions and deliver credential-stealing malware via blockchain-hosted C2 infrastructure.
4 min read
Vulnerabilities Novee Security disclosed Cordyceps, a class of GitHub Actions vulnerabilities exploitable by any free GitHub account. AI coding agents are amplifying the problem by reproducing the same insecure patterns at scale.