How Plugin4Shell Made Reviewed AI Coding Plugins Run Different Code
Four coding agents trusted a commit hash but did not verify the code that Git actually placed in the working tree.
AI Security
Spain’s Reported AI-Agent Breach Happened Behind a Login That Worked
A legitimate login did not mean legitimate intent. Spain's data-protection authority describes an alleged AI-agent attack, while warning that the organisation's report still needs analysis.
The AI Agent Fixed the App by Rewriting the Model Behind It
An AI agent successfully repaired a failing application by changing the shared model behind it. The repair worked, but other behaviour changed too. Who approved the new model?
Can You Defend This Access Decision?
Keep a reviewable record of evidence, limits and decisions.
Millions of Phishing Emails Hid Their Most Suspicious Words in Plain Sight
Millions of messages a day used invisible Unicode characters to split high-signal finance words. Humans saw “funding.” Some security pipelines saw something else.
One Unpatched Langflow Endpoint Could Hand Attackers Your OpenAI and AWS Keys
Attackers are exploiting a critical Langflow endpoint to steal OpenAI keys, AWS credentials and administrator secrets. Patching is only the first step.
Your Password Was Fine. The Stolen Claude Session Still Drained the Account
Infostealers are stealing active Claude sessions, bypassing the need to defeat passwords or multi-factor authentication.
Six AshAI Flaws Turned Prompts Into a Path to Server Takeover
AshAI disclosed six vulnerabilities, including a CVSS 10 flaw that compiled attacker-controlled prompt content as Elixir before any model request.
130 Tech Companies Say We Have Months to Prepare. Their Letter Includes No Deadlines
More than 130 organisations warn that defenders have only months to prepare, but their voluntary pledge includes no deadlines.
Your ‘American’ AI Model May Still Have Chinese DNA
Cisco warns that model publishers and country labels can hide inherited weights, training data and upstream dependencies.