Security Blog

Deep dives into agentic security, MCP exploits, and prompt injection attacks.

LatestResearch

98% Accurate and Still Broken

We built a 98% accurate classifier. Here's why that number doesn't mean what you think it means.

SJSteven·
Newsletter
Stay in the loop
Get our latest agentic security research delivered straight to your inbox.

No spam. Unsubscribe anytime.

#1

The Hidden Risk in Notion 3.0 AI Agents: Web Search Tool Abuse for Data Exfiltration

A critical security vulnerability in Notion 3.0's AI Agents demonstrates how the combination of LLM agents, tool access, and long-term memory creates exploitable attack vectors for data exfiltration.

ARAbi·
#2

Taint Analysis for Agent Tool Calls

Applying classic taint analysis to AI agents: tracing how private or untrusted data flows through tool calls to identify data leaks and tampering risks.

SJSteven·
#3

Heroku Exploit: App Ownership Takeover

A critical vulnerability allows attackers to transfer ownership of a Heroku app by injecting a malicious prompt into its logs.

ARAbi·

More Research

Shopify Exploit: Manipulating Shoppers

A critical vulnerability in Shopify's MCP allows attackers to manipulate consumer purchasing decisions using malicious prompts in product descriptions.

ARAbi·

Neon Exploit: Malicious SQL Injection

Attackers can exploit the Neon MCP server to execute malicious SQL operations. Learn how to protect your database from these threats.

ARAbi·

Azure Exploit: Leaking KeyVault Secrets

A critical vulnerability in Azure's MCP allows attackers to leak KeyVault secrets using malicious prompts.

ARAbi·

Linear Exploit: Bypassing Team ACLs

A critical vulnerability in Linear's MCP allows attackers to bypass team access controls and exfiltrate confidential data using malicious prompts.

ARAbi·

Interested in Partnering?

Looking to collaborate on AI security research, academic projects, or enterprise integrations?