AI SECURITY
NVIDIA Red Team Exposes AI Coding Agent Vulnerability in OpenAI Codex
NVIDIA researchers demonstrate how malicious dependencies can hijack AI coding assistants through AGENTS.md injection, hiding backdoors in pull requests.
NVIDIA Unveils Zero-Trust Architecture for Secure AI Model Deployment
NVIDIA releases open reference architecture for confidential AI factories, enabling secure deployment of proprietary models on shared infrastructure using hardware-backed encryption.
OpenAI Acquires Promptfoo to Bolster Enterprise AI Security Testing
OpenAI announces acquisition of Promptfoo, an AI security platform used by 25% of Fortune 500 companies, to integrate into its Frontier enterprise platform.
OpenAI Codex Security Ditches SAST for AI-Driven Vulnerability Detection
OpenAI explains why Codex Security uses AI constraint reasoning instead of traditional static analysis, aiming to cut false positives in code security scanning.
OpenAI Reveals How ChatGPT Now Fights Prompt Injection Attacks
OpenAI details new 'Safe Url' defense system treating AI prompt injection like social engineering, with attacks succeeding 50% of the time before fixes.
Anthropic AI Discovers 22 Firefox Vulnerabilities in Two Weeks
Claude Opus 4.6 found 14 high-severity Firefox bugs, nearly a fifth of all critical vulnerabilities fixed in 2025. Mozilla shipped fixes to hundreds of millions of users.
OpenAI and Paradigm Launch EVMbench to Test AI Smart Contract Hacking
New benchmark evaluates AI agents' ability to detect, patch, and exploit smart contract vulnerabilities. GPT-5.3-Codex scores 72.2% on exploit tasks.
OpenAI Deploys Web Index Defense Against AI Agent Data Theft
OpenAI reveals new security architecture using independent web indexing to prevent URL-based data exfiltration from ChatGPT and agentic AI systems.
Anthropic Exposes 16M Query Theft Campaign by Chinese AI Labs
Anthropic reveals DeepSeek, Moonshot, and MiniMax ran industrial-scale distillation attacks using 24,000 fake accounts to steal Claude AI capabilities.
Anthropic Launches Claude Code Security to Hunt Zero-Day Vulnerabilities
Anthropic's new Claude Code Security tool found 500+ vulnerabilities in open-source projects. Enterprise and open-source maintainers can apply for early access.
NVIDIA Red Team Releases AI Agent Security Framework Amid Rising Sandbox Threats
NVIDIA's AI Red Team publishes mandatory security controls for AI coding agents, addressing prompt injection attacks and sandbox escape vulnerabilities.
NVIDIA Research Exposes Critical VLM Security Flaws in AI Vision Systems
NVIDIA researchers demonstrate how adversarial image attacks can manipulate vision language models, turning traffic light recognition from 'stop' to 'go' with imperceptible changes.
GitHub's AI Security Protocols: Ensuring Safe and Reliable Agentic Operations
GitHub introduces robust security principles to safeguard AI agents like Copilot, focusing on minimizing risks such as data exfiltration and prompt injection.
Prompt Injection: A Growing Security Concern in AI Systems
Prompt injections are emerging as a significant security challenge for AI systems. Explore how these attacks function and the measures being taken to mitigate their impact.
Meta Introduces Agents Rule of Two for Enhanced AI Security
Meta AI unveils the 'Agents Rule of Two' to mitigate security risks in AI agents, focusing on reducing vulnerabilities such as prompt injection.