#prompt injection
87 stories taggedprompt injection · page 2 of 6.
Your AI email assistant can be fed a fake message while you read a real one
Researchers hid nearly 500 characters of secret instructions inside an ordinary-looking email. The AI summariser obeyed them every single time.
A Single Website Visit Can Poison Your Local AI Agent, Researchers Find
A flaw in Nvidia's NemoClaw lets a malicious webpage secretly rewrite the instructions an AI assistant follows, and the damage survives every conversation that comes after.
Alice Raises $140 Million to Test AI Systems for Weaknesses Before Criminals Find Them
A company that tries to break AI models on purpose, so real attackers cannot, has secured fresh funding to expand that work to more businesses.
The 5% Problem: How a Handful of AI Power Users Became the Riskiest People in Your Company
New Akamai research finds a small group of enthusiastic staff are quietly wiring untested AI tools into serious business systems, and security teams are watching the wrong crowd.
Researchers Find Way to Hide Malicious Instructions Inside Encrypted AI Prompts
A new technique called 'Cryptographic Context Injection' slips harmful commands past the safety filters built into Grok and Gemini by wrapping them in encryption that only the AI unwraps.
Researchers Show How a Booby-Trapped Web Page Can Steal Your Grok Chat Data
Adversa AI's 'Cryptographic Context Injection' technique tricks xAI's Grok into leaking user names, locations and prompts to attacker servers when a user asks it to summarize a page.
Phishing Has a New Problem: The Attacker Isn't Human Anymore
Email defences built for bad links and bad attachments are struggling as AI agents start writing, sending, and even reading the mail on both sides.
Researchers Tricked Microsoft Copilot Into Revealing Its Own Weaknesses, Then Used That Knowledge to Steal Data
A research team at Varonis discovered that simply chatting with Microsoft's AI assistant could expose enough internal detail to build a working attack. Microsoft has patched the flaws, but the technique raises questions that go well beyond one product.
One Click on Copilot Could Have Leaked Your Connected Apps, Researchers Say
Three flaws in Microsoft Copilot Personal, nicknamed CoSnitch, let a booby-trapped link quietly pull data from Gmail, calendars and other services the assistant was connected to.
AI Agents Can Infect Each Other Through Shared Prompt Files, Researchers Show
A preprint from Anthropic and EPFL demonstrates self-spreading instructions jumping between coding agents in a lab setup.
Fortinet Buys AI Security Startup Virtue AI
The cybersecurity giant adds a specialist platform for testing and protecting AI models, chatbots, and autonomous software agents, as the market for AI security tools heats up fast.
How a Rogue Helper Tool Can Trick an AI Coding Assistant Into Leaking Your Secrets
Researchers show that a hostile plugin can smuggle out SSH keys and source code by breaking one big theft into small, innocent-looking steps.
GhostJacking: How Hackers Can Turn an AI Assistant Against Its Own Company
Researchers showed that a single blocked web request, already sitting in a firewall log, was enough to trick an AI agent into handing over a company's entire domain. Here's what that means for organisations using AI tools to manage their systems.
A Single Click Could Have Handed Hackers Your Company's Confluence and Jira Files
Researchers found a flaw in Atlassian's Rovo AI assistant that let an attacker steal data from widely used workplace tools with almost no effort from the victim.
Rovo, Atlassian's AI Assistant, Can Be Tricked Into Leaking Jira and Confluence Data
Two research teams showed how hidden instructions can turn Atlassian's built-in AI helper into a quiet data pipe. Only one of the tricks has been fully closed.