Field notes from the edge.
What our engineers learned this week. Hands-on technical deep-dives, postmortems, and strategy frameworks.
AILiteLLM Vulnerability Chain Lets Low-Privilege Users Take Over AI Gateway Servers
Obsidian Security researchers discovered a critical vulnerability chain in LiteLLM, a widely-used open-source AI gateway, that allows low-privilege accounts to escalate to full admin access and execute arbitrary code on servers. The exploit chains three separate vulnerabilities and could expose all provider API keys and secrets stored on compromised LiteLLM proxy servers, which broker calls to ove
AIThe Beginning of the End of Social Engineering
AI-native operating systems are fundamentally changing cybersecurity defense strategies by embedding intelligence directly into the OS layer to detect and prevent social engineering attacks. This architectural shift transfers the burden of identifying phishing, pretexting, and other manipulation tactics from end users to automated systems, potentially reducing the human error factor that accounts
AIGoogle Sues Chinese Smishing Network Accused of Using Gemini AI in Phishing
Google has filed a lawsuit against a Chinese cybercrime network for allegedly weaponizing its Gemini AI to facilitate SMS phishing (smishing) attacks targeting Americans. The network reportedly operates a phishing-as-a-service (PhaaS) platform called Outsider, demonstrating how threat actors are exploiting generative AI tools for malicious purposes.
AIClaude Fable 5 Doesn't Change the Mythos Security Story
Anthropic has released Claude Fable 5, positioning it as a production-ready version of their Mythos model that has been optimized for general enterprise use. Mythos 5 represents an incremental upgrade over the preview version, while Fable 5 focuses on safety and stability for broader deployment. The release maintains continuity in Anthropic's security approach rather than introducing fundamental c
AIAgentjacking Attack Tricks AI Coding Agents Into Running Malicious Code
Cybersecurity researchers at Tenet Security have identified a new attack vector called 'Agentjacking' that exploits AI coding agents to execute malicious code on developer systems. The attack leverages fake error reports through platforms like Sentry, an open-source error-tracking tool, to trick AI agents into running arbitrary code.
AILangGraph Flaw Chain Exposes Self-Hosted AI Agents to Remote Code Execution
Three security vulnerabilities have been discovered and patched in LangGraph, an open-source framework for building multi-agent AI applications. The most critical flaw involves a vulnerability chain that could enable remote code execution, with an SQL injection identified as one of the attack vectors. Organizations using self-hosted LangGraph deployments for AI agent development should prioritize
AINew Attacks Trick OpenClaw AI Agent Into Running Code and Leaking Secrets
Security researchers from Imperva and Varonis have independently demonstrated critical vulnerabilities in OpenClaw, a widely-used self-hosted AI agent, showing it can be manipulated to execute malicious code and expose sensitive information. The attacks exploit the agent's processing of seemingly benign inputs like vCards, shared contacts, and location data, allowing attackers to embed hidden inst
AIThreatsDay Bulletin: Worm Code Leaked, AI Agent Phished, Claude Action Patch + 28 New Stories
The cybersecurity landscape is evolving with increasingly sophisticated threats, including publicly leaked supply chain attack tools, premium remote access trojans with browser-cloning capabilities, and demonstrated vulnerabilities in AI agents that can be exploited to extract credentials. The professionalization of cybercrime infrastructure, with mule networks operating as polished SaaS-like serv
AIAI Broke Vulnerability Management. That's Why CISOs Are Moving Budget to BAS.
Traditional vulnerability management relied on a time buffer between vulnerability discovery and exploitation, allowing teams to triage and patch systematically. AI has eliminated this buffer by accelerating threat actors' ability to weaponize vulnerabilities, rendering conventional approaches obsolete. This shift is driving CISOs to reallocate budgets toward Breach and Attack Simulation (BAS) sol
AIUnpatched Langflow Flaw CVE-2026-5027 Exploited for Unauthenticated RCE
An unpatched high-severity vulnerability (CVE-2026-5027) in Langflow, an open-source low-code AI application development platform, is being actively exploited in the wild. The path traversal flaw, with a CVSS score of 8.8, enables unauthenticated attackers to achieve remote code execution by writing files to arbitrary locations on affected systems.
AIA Record-Breaking Patch Tuesday for June 2026
Microsoft released a record-breaking 200 security patches in June 2026's Patch Tuesday, with nearly three dozen rated critical and at least three zero-days publicly exploitable. The unprecedented volume is attributed to increased AI-powered vulnerability discovery by both Microsoft engineers and security researchers, a trend expected to continue. The release was complicated by ongoing tensions wit
AIDefend against frontier cyber models: Cloudflare's architecture as customer zero
Cloudflare details its defense architecture against AI-powered cyber threats, emphasizing that architectural design matters more than patching speed when facing frontier AI models like Mythos. The company operates as 'customer zero' for its own security products, using its visibility into ~20% of global web traffic to detect and block threats in real-time through integrated WAF and threat intellig
AIAnthropic’s Project Glasswing Update
Anthropic's Project Glasswing, launched in April to help companies identify software vulnerabilities using AI, has produced questionable results despite positive media coverage. While the project claims to find numerous vulnerabilities, very few have actually been patched, and Anthropic's refusal to release detailed data raises transparency concerns about the initiative's actual effectiveness.
AINew ChatGPT Lockdown Mode Limits Tools That Could Enable Data Exfiltration
OpenAI has introduced Lockdown Mode for ChatGPT personal accounts to mitigate data exfiltration risks from prompt injection attacks. The feature targets users and organizations handling sensitive data by restricting certain tools and capabilities. It is available across all personal account tiers including Free, Go, Plus, and Pro.
AIAI Agent Uncovers 21 Zero-Days in FFmpeg; Chrome Patches Record 429 Bugs
A security startup's autonomous AI agent discovered 21 zero-day vulnerabilities in FFmpeg, a widely-used media library embedded in countless video applications. This AI-driven discovery coincided with Google's Chrome 149 release, which patched a record 429 security bugs—though notably, only the FFmpeg vulnerabilities were identified through AI methods.
AIOnly 10% of SOCs Say They’re Getting Excellent Value From AI. Here’s What the Second Wave Has to Deliver
AI in Security Operations Centers (SOCs) has rapidly evolved from a marketing concept to a significant budget priority, with billions being invested in AI-powered security platforms and tools. Despite widespread adoption and deployment, only 10% of SOCs report receiving excellent value from their AI investments, indicating a gap between implementation and realized benefits. The industry is now ent
AITrump AI Order Seeks Voluntary Frontier Model Testing
The White House has issued an executive order creating a voluntary framework that allows early government access to frontier AI models for testing and evaluation. The order also includes provisions for increased federal investment in AI security infrastructure and oversight capabilities.
AIThreatsDay Bulletin: AI Agents Gone Wrong, Sketchy C2 Tools, ClickFix Tricks, JS Backdoors & 20+ New Stories
The cybersecurity landscape continues to deteriorate with persistent vulnerabilities in plugins, legacy systems, and trusted applications being exploited through increasingly sophisticated attack vectors. AI-enabled threats are now causing tangible system failures while threat actors gain access to more advanced tooling at lower costs. The security infrastructure remains fragile, with compromised
AIHacking Meta’s AI Chatbot
Hackers exploited Meta's AI support chatbot to hijack Instagram accounts by using VPNs to spoof locations and convincing the bot to add new email addresses and reset passwords. While Meta claims the specific vulnerability has been patched, the incident highlights a fundamental security concern: LLM-based chatbots lack the trustworthiness required for sensitive account management functions, and sim