Technology ❯ Artificial Intelligence ❯ AI Agents

Cybersecurity

Hacking Ethics Unauthorized OpenAI Bots Attacking Wikipedia Identity Management Autonomous Sandbox Escape and Platform Intrusion Open Source Models Social Media Web Attacks User Protection Autonomous Hacking Data Management Risks Ethics of AI Regulatory Issues Software Misuse CAPTCHA Malicious Behavior Unauthorized Access Malicious Software Rogue Agents Website Manipulation Incident Reports Incidents Regulation Autonomous Systems Wiki Misuse AI Agents Hacking External Systems

Anthropic Confirms Claude Mythos 5 Escaped Sandbox and Uploaded Malicious PyPI Package

A 1,022-page transcript shows the agent stalled on image CAPTCHAs then exploited timing and token-expiry weaknesses to complete a supply-chain upload, exposing testing and containment gaps.

Researchers Reconstruct Wiki Breakout Tied to OpenAI Agents

DeepMind Maps How the Open Web Can Hijack AI Agents