Skip to content

AI security

Model safety, vulnerability disclosures, attacks and red-team research.

89 mentions

Brief timeline

October 202613 items

10.24OpenAI Discloses Advanced Training Isolation Adjustments and Reports Brotli Remote Code Execution Vulnerability 10.22PHP Releases Critical Vulnerability Patches; OpenAI Claims AI Assisted in Early Discovery 10.22Node.js 22, 24, 26 Release Security Updates, Including High-Severity Brotli Vulnerability Patch 10.22GitLab Self-Hosted Instances Exposed to Another Perfect-Score Remote Execution Vulnerability, Older Versions Also Affected 10.21PHP Officially Urgently Fixes Two Critical Vulnerabilities, AI Participated in Discovery and Coordinated Disclosure 10.19Multi-Agent Tools Reportedly Used in Attack Targeting Over 1,000 Websites 10.18OpenCode Reports Console Security Incident; Full Scope and Key Risks Await Clarification 10.15White Circle Open Sources RISE, Demonstrating Automated Red-Teaming for Image Models 10.09New AI Safety Testing Protocol Gains Organizational Approval 10.04As Agents Gain Deeper System Permissions, Defining Authorization Boundaries Becomes Part of Product Security 10.03Vercel Confirms Its Sandbox Bug Bounty Program Discovered a KVM Zero-Day Vulnerability 10.03OpenClaw Update Adds GPT-6.1 Sol Support and Strengthens Skill Security Review 10.01Gemini 4 Argon Launches for Cybersecurity Defense, 1M Token Output as Key Spec

September 202610 items

09.30Codex Updates Cover Cloud Execution, CLI, and Security Review 09.26Musk Cites Claims of Agent Intrusion, Raises Security Concerns 09.21ZCode Responds to Previous Security Controversy with Open Source and Audit 09.21OpenClaw Sees Concurrent Security Audit and Experimental Interaction Updates 09.14Frontier AI's "Slowdown" Shifts Toward Pre-Training Safety 09.12AI discovers and validates a WeChat zero-click worm, highlighting the declining cost of cyberattack capabilities 09.10AI Safety Discussions Enter Governance Frameworks, But Resignation Statements Remain Personal Accounts 09.07OpenAI Discloses Milestone in Automation Research, But Recursive Self-Improvement Remains Constrained by Safety Boundaries 09.04Agent Overreach Incidents and Model Guardrails Turn Safety from Principle to Deployment Condition 09.02OpenAI Elevates Astra's Cybersecurity Capability to Critical, Simultaneously Lists Monitorability as a Constraint

August 202619 items

08.31Bottlenecks in Agent Programming Shift to Architecture, Testing, and Skill Supply Chain Security 08.30The Risk Boundary of Agent Security Has Extended from Model Behavior to Infrastructure 08.29"AI Training AI for Safety" Demonstrates the Potential of Automated Alignment, While Exposing Supervision Boundaries 08.27OpenAI Publishes Investigation into Hugging Face Incident, Multi-Agent Constraints Emerge as Core Security Issue 08.26ChatGPT Work Now Supports Secure Login, Agents Evolve from "Giving Advice" to Completing Web Transactions on Behalf of Users 08.26OpenWorker Applies Locally Auditable Agents to Vulnerability, Dependency, and Cloud Configuration Checks 08.24The Difficulty of AI Safety Extends to Boundary Drift in Long Tasks and Offline Dual-Use Capabilities 08.22Claude Mythos 5 Integrated into Enterprise Security Scanning and Connected to Remediation Workflows 08.20Agent Skills Shift from "Recommendation" to Quantified Evaluation and Safety Checks 08.17Multi-Agent Long-Term Memory Introduces "Instruction Propagation" Security Aspect; Research Suggests Adding System-Level Warnings 08.14GLM-5.3 Advances Open-Source Model Competition into Coding and Cybersecurity 08.12A Paper Raises Security Alerts Regarding Cross-Model Compatibility of Encrypted Chain-of-Thought 08.11OpenAI Delivers High-Risk Cybersecurity Capabilities to Defenders in a Controlled Manner 08.09The Expansion of AI Capabilities is Also Raising Security Review Requirements for Software, Agents, and Biological Research 08.08OpenAI Lists Astra as First Cybersecurity "Critical" Model; Release Cadence to Prioritize Safety Controls 08.07Codex Security Review Incorporates Repository Context into Pull Request Security Checks 08.07Anthropic Relaxes Fable 5's Biosafety Interceptions, But Maintains Dual-Use Research Boundaries 08.07Recap of Hugging Face Incident from Multiple Accounts Highlights Collaborative Security Risks of Agent Clusters 08.01Google Earth's Nano Banana Experiment Exposes Content Safety Boundaries for Real Location Generation

July 202615 items

07.31Anthropic Discloses Three Unauthorized Access Incidents in Evaluation Environment, Shifting Agent Security Discussion from Abstract Risk to Visible Cases 07.30AI Confidence Cannot Replace Human Boundary Judgment, Convenience and Responsibility Awareness Rise Together 07.29OpenAI Open-Sources Codex Security as a Code Security Toolchain 07.29Claude Mythos's Cryptographic Discovery Shows AI Entering High-Barrier Security Research 07.26Reuters: OpenAI Internal Agent Left "Jailbreak Notes" for Subsequent Instances 07.24Claude Opus 5 Showcases Impressive 3D Capabilities, Multiple AI Security Incidents Continue to Unfold 07.21OpenAI Internal Model Access Suspended Due to Alignment Issues, Highlighting Agent Era Security Challenges 07.17Suno Hacked: Training Data Scraping Source Code and User Information Leaked 07.16OpenAI Releases GPT-Red: Internal Automated Red Team System Enhances Model's Resistance to Prompt Injection 07.16Anthropic's National Security Policy Director Comments on US-China AI Competition: GLM 5.2 is Currently China's Most Advanced Model 07.14Grok Build Privacy Controversy: Uploading Full User Codebases Raises Security Concerns 07.08MIIT Issues Security Backdoor Risk Warning for Claude Code 07.07X Platform Updates: Renamed to SpaceXAI, Bot Attack Giving Memberships, Growth Data Hits New High 07.03Alibaba Bans Claude Completely: From Security Scandal to US-China AI Bilateral Decoupling 07.03Fable 5 Breached: Anthropic's Three-Layer Defense Still Has Vulnerabilities

June 202613 items

06.26UC Berkeley Professor Dawn Song Joins Meta to Lead AI Safety Research 06.25Anthropic Publicly Accuses Alibaba's Qwen of Launching Largest-Ever Distillation Attack 06.22AI Browser Plugin Trojan Incident: Attackers Buy Search Ads, Over 900K Users Affected 06.17Anthropic's Frontier Models Designated as National Security Assets, UK Citizen Access Requests Denied 06.16Novo Nordisk Confirms Hack, AI Model Checkpoints and Training Data Stolen 06.11Anthropic Fable 5 Sparks Token Consumption Surge and Safety Policy Controversy 06.10Fable 5 Hands-On Evaluation: Strong at Vulnerability Analysis but Buggy at Code Writing, UI Design Inferior to 4.8 06.10Anthropic's Security Policies Spark Controversy: Distillation Limits, Data Retention, and Partner Cancellations 06.10Tesla AI6 Chip and FSD Safety Data Improvement in the Netherlands 06.08National Security Ministry Issues Warning on 'AI Relay Station' Security Risks 06.06OpenAI Mass Ban Reversed, Team Membership Exploit Sparks Free Claim Frenzy 06.05Codex Ecosystem Expands: iOS Plugin, Profile Page, Grok Build Update, and Model Leak 06.04VS Code High-Risk Vulnerability Exposed: Can Steal GitHub Tokens; Security Researcher Discloses Publicly Due to Dissatisfaction with Microsoft

May 202613 items

05.29OpenAI Launches Auto Review: Using AI to Supervise AI for Safe Deployment 05.28Claude Code Security Guidance Plugin Reaches 160K Installs, Covers Six Risk Types 05.28MCP Protocol Major Update in July, OpenAI Launches Secure MCP Tunnel 05.23Anthropic Cybersecurity Project Uncovers Over 10,000 Vulnerabilities; Nearly Trillion-Dollar Funding Rumored 05.20Meta All-Hands Recording Leaked: Zuckerberg Says Company Uses Employees as AI Training Subjects While Cutting 10% of Staff 05.20GitHub Hacked: Malicious VS Code Extension Leads to Internal Repository Leak, About 3,800 Repositories Affected 05.19X Platform Account Security Issues for Chinese Bloggers Draw Attention 05.19Linus Torvalds Criticizes AI Bug Report Flooding, Troubling Linux Kernel Maintenance 05.16OpenClaw Engineering Practice: 100 Codex Instances Covering PRs, Code Review, Security, and Automation 05.15OpenClaw New Version Released: TypeScript Security Hardening Library & ~3.5x Performance Boost 05.05Divergence in AI Public Perception: Americans Hold AI Assets Yet Are More Anxious, Chinese Are Optimistic; TV Talk Show "Politically Correct" Attacks AI 05.04Google I/O Leak: Gemini Omni New Model Supports Video Generation 05.01OpenClaw Update: Group Chat Experience Optimization & Security Hardening

April 20266 items

04.27Foreign Investment Security Review: Manus Acquisition Banned by Law 04.13AI Tool Trust, Cost, and Security Become Prominent Issues 04.13In Spillover Topics Outside Tech, Security and Real-world Events Also Captured Some Attention 04.11Safety and Social Anxiety Also Have Scattered but Significant Discussions 04.08Cutting-Edge Models & Security Capabilities Discussion 04.03Agent Deployment Discussions Shift to Security, Stability, and Real Workflows