Claude Code Auto Mode: Why a 0% Lab Result Became 80% in the Wild
A researcher hit 60-80% prompt injection success on Claude Code after Anthropic benchmarked 0.00%. The gap reveals where the real security boundary sits.
When Prompt Injection Goes to Court: The Connecticut Filing That Changed the Conversation
A plaintiff embedded invisible AI instructions in court filings to manipulate automated document review — and got caught. The case shows how prompt injection moves from developer concern to legal reality.
Researchers Can Now Reconstruct LLM Prompts From Outputs With Near-Perfect Accuracy
IIT Bombay and Adobe Research's 'Previous-Token Prediction' technique extracts system prompts from model outputs without access to weights. The IP protection implications for enterprises deploying proprietary AI are significant.
Atlassian Rovo: The PDF Flaw Leaking Jira Data to Attackers
Hidden PDF text can hijack Atlassian Rovo and silently exfiltrate Jira tickets and Confluence data. Reported in May. Unpatched as of August 2026.
The First Self-Replicating Prompt Injection Worm Targets Word
A new attack hides invisible instructions in Word documents that make Copilot alter content and copy itself into every new document it generates next.
Opus 5 Just Posted a 0% Rate on the Attack That Breaks AI Agents
Anthropic's Opus 5 scored 0% success across 129 browser-based prompt injection tests with Auto Mode on — a benchmark OpenAI called potentially unsolvable.
Hugging Face Was Hacked by an AI Agent — Then Fought Back With One Too
An autonomous AI agent breached Hugging Face's production infrastructure using 17,000 recorded actions. The response revealed a critical gap in every team's AI security posture.
Claude Code's Verification Gap: When Agents Run Code Without Looking
Security researchers caught Claude Code executing a hidden DNS payload without user preview — handing attackers full machine access via a GitHub link.
6,000 Attempts to Hack an AI Assistant. Zero Succeeded.
2,000 hackers. 6,000 prompt injection attempts. Zero secrets leaked. What the 30-day hackmyclaw.com challenge tells us about the current state of AI security hardening.
Prompt Injection as Role Confusion: A New Framework for an Old Problem
A new paper argues that prompt injection succeeds because models prioritize text style over semantic role — and shows a defense (destyling) that reduces attack success from 61% to 10%.
ChatGPT's Lockdown Mode Blocks Data Exfiltration—But Not the Injection Itself
OpenAI's new Lockdown Mode disables web access, deep research, agent mode, and network calls for Canvas-generated code—stopping the last mile of prompt injection attacks while acknowledging the underlying problem remains unsolved.
AI Turns Security Patches Into Working Exploits in 30 Minutes. The 90-Day Window Is Over.
An LLM turned a React security patch into a working exploit in 30 minutes — collapsing the assumption behind every coordinated disclosure timeline.
An AI Agent Hacked McKinsey's Platform in Two Hours Using SQL Injection
Security firm Codewall turned an offensive AI agent loose on Lilli, McKinsey's internal AI platform. In two hours, it had read/write access to 46.5M chat messages — via a decades-old bug.