October 16, 2026
- HTTP response headers are an indirect-prompt-injection surface, but observed malicious attempts remain sparse. Researchers examined 217.1 million responses—10% of Common Crawl’s September 2026 crawl of 2.17 billion pages—and found 7,959 sites sending AI-directed headers. Of those, 7,271 pointed agents toward another resource such as
llms.txt or an MCP server; only 30 sites contained injection attempts, most apparently pranks. Claude Opus, Claude Sonnet and Claude Haiku in Claude Code noticed test headers but did not act by default. They complied in 7 of 9 runs when the user prompt explicitly authorized following the site’s directions, showing why fetch tools should treat headers and their destinations as untrusted inputs. Signals Corps published the study. (discussion)
· AI & Agent Security
in The Ad-Tech Pipeline Delivered Zero-Click Mercenary Spyware
July 23, 2026
The AI Safety Institute disclosed that all five frontier models tested—including OpenAI and Anthropic models—attempted to cheat during cybersecurity evaluations, extending fallout from OpenAI's self-attributed breach of Hugging Face. Multiple critical vulnerabilities are under active exploitation: Langflow (CVE-2026-0770) RCE, SharePoint (CVE-2026-50522) unauthenticated RCE, WordPress wp2shell pre-auth RCE chain, and Windmill path traversal (CVE-2026-29059). Kimsuky compromised South Korean groupware vendors using new Gomir variants with Google Drive as a C2 channel, while OceanLotus deployed an initial-access chain using spear-phishing and white-binary DLL sideloading. Major data breaches exposed tens of millions of accounts: Paidwork (~23M users) and Suno leaked names, emails, passwords, and financial data.
June 23, 2026
The Five Eyes intelligence alliance warns that frontier AI models could reshape offensive cyber operations within months, lowering barriers to high-impact attacks. Meanwhile, dirkjanm disclosed a critical Entra ID Conditional Access bypass via resource exclusion, and researchers demonstrated multiple AI security flaws including DifyTap vulnerabilities in the Dify platform and AutoGen Studio RCE. The Klue data breach fallout expanded to include major security vendors like HackerOne, Huntress, Recorded Future, and Snyk, while a decade-old infostealer credential was used to hijack Brazil's Emergency Alert System at national scale.