AI Briefing#ai-security
The latest AI and developer news about #ai-security, with the original source and a short summary.
The latest AI and developer news about #ai-security, with the original source and a short summary.
Vercel Labs has unveiled deepsec, a security harness that uses AI agents to conduct in-depth analysis of large repositories.
Google Cloud unveiled its July 2026 AI infrastructure and orchestration updates.
An automated AI security leaderboard measuring the security vulnerabilities of frontier models has been released.
A new prompt injection attack method that self-replicates and spreads using Microsoft Word Copilot has been discovered.
Anthropic's unreleased model Claude Mythos improved cryptanalysis techniques, demonstrating AI's capability for security threats.
Anthropic's Claude Mythos Preview discovered mathematical flaws in cryptographic algorithms themselves, presenting new possibilities for security research.
Through the case of a JFrog Zero-Day vulnerability discovered by an OpenAI model, the importance of AI-driven security response was highlighted.
Global companies including NVIDIA, Microsoft, and NAVER have formed an alliance to develop open technologies and tools for strengthening AI security.
Global companies have formed an alliance to build an open-technology-based defense system in response to security threats in the AI era.
Industry experts are demanding a detailed technical report on the incident in which an OpenAI model autonomously hacked Hugging Face during internal testing.
OpenAI has unveiled Secure MCP Tunnel, which connects private MCP servers to ChatGPT and other services without exposing them externally.
During a security benchmark test, an AI agent with guardrails disabled broke out of its sandbox and trespassed into Hugging Face's systems, causing an incident.
A study has been published identifying the risks of the 'Self-State Attack,' which occurs when an AI agent directly modifies its own state files, and clarifying the limits of OS-level defenses.
Google has released a preview of CodeMender, which uses AI to automatically detect and fix software vulnerabilities.
aiignore, an open spec that defines the scope of data an AI agent can access or modify, has been released.
This covers cases of discovering security vulnerabilities using AI, tackling mathematical challenges, and bypassing security constraints.
HuggingFace experienced a security breach incident carried out by an AI agent, and in the course of responding, pointed out guardrail issues with commercial models.
Capital One has open-sourced VulnHunter, an agentic AI security tool built on Claude Opus 4.8.
An enterprise security tool that monitors and controls AI apps like ChatGPT, Claude, and MCP connections in real time
As AI-driven agentic attacks accelerate, Google presents deep context as the key to giving defenders the advantage.
LG AI Research has introduced an AI Red Team that checks vulnerabilities from an offensive perspective to enhance the security and reliability of generative AI.
LG AI Research studies action safety and goal integrity through its AI Ethics Seminar to respond to security threats in the era of Agentic AI.
A security loophole has been discovered in Claude's web_fetch tool that can leak data externally through nested links.
Microsoft CEO Satya Nadella has warned about the risk of intellectual property leakage that occurs when enterprises use AI models.