ai-safety 55
- Anthropic Releases Claude Opus 5.5 With Stricter Cybersecurity Safeguards
- UK NCSC Official: AI Set to Help Cyberattackers More Than Defenders
- Cisco Talos Details CLOSEDQUORUM, First Reported Autonomous AI C2 Implant
- OpenAI, Anthropic, Google, and 100+ Companies Call for Action to Defend Against Rogue AI
- AWS Extends Bedrock Guardrails to Tool Interactions via Strands Agents SDK
- Amazon Kiro Prompt Injection Flaw Can Exfiltrate Data via Kiro Powers
- Google DeepMind Pilots the World's First Double-Blind AI Evaluations
- OpenAI Says Agents Coordinated via Makeshift Message Board Ahead of Hugging Face Hack
- Researcher Claims Control of ChatGPT Secure Sandbox
- Route Amazon Bedrock Guardrails Interventions to Amazon Security Lake
- OpenAI Finds More Evidence of Agents 'Running Amok' in Hugging Face Incident
- OpenAI Disrupts Cambodia-Based AI Scam Operation
- Anthropic Says Claude Models Autonomously Hacked Three Real Organizations
- Okta Buys AI Security Startup Permiso for About $200M
- 'DangleGeddon': Researchers Warn AI Could Weaponize Dangling DNS Records at Scale
- Chinese-Speaking Threat Actor Harnesses AI Models for Autonomous Cyberattacks
- OpenAI's Rogue Agent Breach Widens: JFrog Zero-Days and Stolen Credentials Hit Hugging Face and Others
- xAI Open-Sources Grok CLI After It Silently Uploaded Users' Files to the Cloud
- AI Agent Runs Ransomware Attack Start to Finish via Langflow RCE
- Critical Cursor Flaws Let Prompt Injection Escape Sandbox
- AI-Generated Browser Ransomware Abuses Chromium API
- Microsoft Fixes AutoGen Studio Flaw That Enabled Code Execution
- Amazon Security Research Reportedly Behind Anthropic Fable Export Ban
- Anthropic Releases Claude Fable 5: First Mythos-Class Model Available to the Public
- Google Launches AI Threat Defense with Automated Security Operations Agents
- OpenAI Rolls Out ChatGPT Lockdown Mode to Limit Prompt Injection Data Exfiltration
- OpenAI Launches Lockdown Mode to Limit Data Exfiltration from Prompt Injection Attacks
- Trail of Bits: Malicious AI Skills Bypass Every Public Skill Scanner
- AI-Built Ransomware Toolkit Automates EDR Evasion and Active Directory Discovery
- Microsoft Introduces Portable Policy Files for Defining and Enforcing AI Agent Behavior
- Google Deploys AI Deepfake Call Detection to Flag Impersonation Scams in Phone App
- Anthropic Expands Mythos Security Program to 150 Organizations Across Critical Infrastructure
- Meta AI Confused Deputy Flaw Lets Attackers Take Over Instagram Accounts
- Meta AI Confused Deputy Flaw Enables High-Profile Instagram Account Takeovers
- SymJack Attack Weaponizes AI Coding Agents as Supply Chain Delivery Systems
- Attackers Use AI Chatbot Responses to Deliver Cryptojacking Malware
- Anthropic's Project Glasswing Uncovers 10,000 High-Severity Vulnerabilities in Critical Software
- Langflow CVE-2025-34291 (CVSS 9.4) Added to CISA KEV Under Active Exploitation
- Pwn2Own Berlin 2026: $1.3M Paid for 47 Zero-Days in Windows, Linux, VMware, and AI Products
- Hugging Face Model Tokenizer Files Can Be Weaponized to Hijack Outputs and Exfiltrate Data
- Claude AI Guided Attackers to OT Assets During Water Utility Intrusion
- Researchers Bypass Claude Safety Guardrails via 'Gaslighting' Technique
- Elon Musk Testifies xAI Trained Grok on OpenAI Model Outputs, Highlighting Distillation Dispute
- OpenAI Publishes Root Cause Analysis of GPT-5 Goblin Behavior Quirks
- AI-Generated Celebrity Deepfakes Drive Scam Campaigns on TikTok
- Claude Mythos Accelerates Vulnerability Discovery—but Remediation Teams Aren't Keeping Pace
- Cisco Discovers Memory Vulnerability in Anthropic AI Agent Framework
- Claude Mythos Preview Found 271 Firefox Vulnerabilities in Anthropic-Mozilla Collaboration
- Unauthorized Group Claimed to Have Gained Access to Anthropic's Restricted Mythos Cyber AI
- Google Antigravity AI IDE: Prompt Injection Chained to Sandbox Escape and Code Execution
- OpenAI Launches GPT-5.4-Cyber, a Fine-Tuned Model for Defensive Security Use Cases
- Anthropic Co-Founder Confirms Briefing Trump Administration on Mythos Despite Active Lawsuit
- Anthropic Restricts Mythos Preview After Model Autonomously Exploits Zero-Days in Major OS and Browsers
- DoD Flags Anthropic as Supply-Chain Risk While Trump Officials Push Banks to Test Mythos
- Apple Intelligence Guardrails Bypassed via Neural Exect and Unicode Manipulation