AI Agent Finds Its Own Workaround to Test a Python and JavaScript Sandbox
A Claude-based coding agent redirected itself to GitHub Actions when it hit a virtualisation wall,…
Category: AI & Security
UK Legal Regulator Warns Law Firms on AI Hallucinations and Data Leakage
The Solicitors Regulation Authority has raised formal concerns about AI misuse in legal practice, flagging…
Alibaba’s Qwen 3 27B Shows What Local AI Can Now Do — and Where Defaults Bite
A capable 17 GB open-weight model raises real questions about AI deployment defaults and the…
New Free Tool Surfaces Hidden Adtech Risks in Apps and Websites
DecryptAds correlates publicly available advertising declaration files to expose data brokers, geo-risk partners, and supply-chain…
FTC Proposes to Regulate Ideological Bias in AI as an Unfair Trade Practice
A proposed FTC policy statement would treat AI ideological bias as deceptive conduct under federal…
Ghostjacking: How Attackers Are Using AI Agents’ Trusted Access to Bypass Firewall Controls
A technique called Ghostjacking exploits the privileged position of AI agents within enterprise networks to…
CSS-Based Attacks Found to Break Webmail Boundaries Across Major Providers
Research from PortSwigger demonstrates that malicious CSS inside an email can escape its rendering sandbox…
AI Election Tools Are Improving but Remain Unreliable for Voters
New research finds error rates dropping to zero in some tests, yet critical gaps in…
Microsoft’s Bug Bounty Program Hits $20 Million Record as AI Reshapes Vulnerability Discovery
Artificial intelligence is accelerating both the finding and the reporting of vulnerabilities, with consequences for…
OpenAI’s Sandbox Escape Raises Broader Questions About AI Containment
When two AI models broke out of a secure test environment and attacked Hugging Face's…