Tag: Prompt Injection
An accounting bot blew $50,000 in cloud charges in an hour
Mandiant's enterprise AI report finds runaway agents, prompt injection, and weak access controls already costing real money.
Planted prompt turned ChatGPT into a hidden Gmail relay
Check Point shows a planted instruction can make ChatGPT exfiltrate Gmail data through a covert channel between accounts.
Russia-aligned group trips AI scanners with nuke prompt
ESET says Russia-aligned UAC-0099 hides a nuclear weapon request in malware comments to stall AI-powered triage.
Amazon’s Kiro IDE leaks data when poisoned projects get opened
A prompt injection flaw in Amazon's Kiro IDE can push sensitive local data to attackers when a poisoned project is…
Encrypted webpages trick Grok into leaking chat data to attackers
Adversa AI shows how ciphertext hidden in a page can make xAI's Grok hand over chat details with no warning.
Copilot reveals the trick that lets one click drain connected apps
Varonis shows how one click on a crafted link can drain data from connected apps in Microsoft Copilot Personal.
Poisoned logs turn trusted platforms into agent hijackers
Tenet Security's Ghostjacking demo shows attackers planting instructions in logs that AI agents read and execute, targeting Cloudflare, Datadog, and…
One-click RovoBlast exploit leaks data from Atlassian workspaces
Two firms found independent ways to turn Atlassian's Rovo assistant into a one-click data leak.