Cybersecurity news — data breaches, vulnerability disclosures, AI-powered security tools, threat intelligence, and enterprise security research.
Manifold Security has added security scores for over 7,700 MCP (Model Context Protocol) servers to its supply chain intelligence platform, highlighting a critical security gap. It warns that the rapid adoption of MCP ser…
Users often unknowingly expose sensitive information—such as proprietary code or client requirements—to AI tools without understanding where their conversation data ends up, including backups, training data, and third-pa…
Dominion Observatory, a system that provides behavioral trust scores for over 14,800 MCP servers by analyzing their runtime behavior rather than static source code. Unlike traditional static analysis, which only checks f…
The open-source tool Promptfoo, acquired by OpenAI in March 2026, maps its 155 attack plugins to the OWASP LLM Top 10 2025 list for structured red teaming of LLM-powered products. It details the 2025 revision's new categ…
Practical guide for engineers and security teams evaluating whether to run large AI models locally, in private cloud, or via secure enterprise platforms. It argues that AI performance depends not only on GPU power but al…
The author was scammed by an API relay provider that substituted GPT-3.5 for GPT-4, inflated token counts, and injected hidden prompts. In response, they built API DNA, a free tool that performs quick and deep scans to d…
Against using admin-level database credentials for AI agents, as this poses significant security risks. Instead, it recommends granting each AI agent a narrowly scoped credential tailored to its specific task, with read-…
The ClawHavoc campaign of January–February 2026 exploited a trust assumption in an AI-agent skill marketplace, using 1,184 malicious skills to infect approximately 300,000 users with a macOS infostealer. In response, the…
Prompt injection is an architectural design flaw in LLMs where the data and control channels share the same input, making it nearly impossible to fully eradicate. The guide analyzes four common attack techniques and outl…
AgentThreatBench is a new security benchmark that operationalizes the OWASP Top 10 for Agentic Applications (2026) into executable tasks, designed to measure whether an AI agent can be hijacked by malicious inputs in its…
To defend against OWASP ASI06 memory poisoning attacks in the OpenAI Agents SDK by using Pydantic's `@field_validator` to validate agent context data. It demonstrates how to integrate the OWASP Agent Memory Guard library…
Codev's multi-model review system caught two distinct bugs that no single AI model would have identified alone: Codex detected a Unix socket permission flaw (missed by Claude and Gemini), while Claude spotted an OAuth no…
Tool that removes both visible and invisible AI watermarks from images generated by models like Google Gemini, ChatGPT, and Midjourney. It strips metadata such as SynthID, C2PA credentials, and EXIF labels, and offers fe…
A maximum-severity vulnerability, tracked as CVE-2026-45829, has been discovered in the Python FastAPI version of the open-source AI database ChromaDB, allowing unauthenticated attackers to execute arbitrary code on expo…
Shay Shwartz, a former teen hacker turned cybersecurity expert who worked on Israel's Iron Dome project, has raised $28 million for his startup Ocean, an agentic email security platform designed to combat AI-powered phis…
Based solely on the provided text, the article's title indicates that Google is promoting its increased capital expenditures and "tokenmaxxing" (likely a reference to cryptocurrency or token-related initiatives) during a…
Anthropic has released two new enterprise features for its Claude Managed Agents platform: self-hosted sandboxes (public beta) and MCP tunnels (research preview). Self-hosted sandboxes allow tool execution to run on cust…
A security researcher discovered that the U.S. Cybersecurity & Infrastructure Agency (CISA) had exposed a large cache of plaintext passwords, SSH private keys, and other sensitive credentials in a public GitHub repositor…
At Google I/O, the company announced wider external availability of CodeMender, an AI agent for code security that flags and fixes vulnerabilities, positioning it as a tool to help secure global code bases. This move fol…
OpenBSD 7.9 has been released, introducing delayed hibernation support on amd64 and improved handling of low disk space during system upgrades. The release also adds VA-API and open Widevine support to its Chromium port,…