Notion Deploys Custom AI Agents
Notion is rolling out "custom agents" that allow users to automate workflows like report generation and database updates using natural language. While a productivity feature, the trend of AI agents as practical "teammates" is expected to be mirrored in security platforms for tasks like alert triage and vulnerability scanning.
Notion's custom agents, which launched in public beta on February 24, 2026, represent a shift from on-demand AI chats to autonomous background workers. During the initial testing phase, users built over 21,000 agents, while Notion itself now runs approximately 2,800 internally—more agents than employees. These agents operate on triggers from connected apps like Slack, Gmail, GitHub, and Figma, allowing them to perform cross-platform workflows such as routing engineering tickets or compiling sales reports. Following a free trial period ending May 3, 2026, the feature will shift to a usage-based model, costing $10 per 1,000 credits, separate from Notion's primary AI add-on. This trend of agentic AI is mirrored in offensive security, where it's being applied to automate penetration testing. Projects like XBOW and CAI are pioneering agents that can independently probe systems, discover vulnerabilities, and craft exploits, with some systems solving Capture The Flag (CTF) challenges thousands of times faster than humans. This allows for continuous, scalable security validation that surpasses the speed of manual testing. However, the power of these agents introduces new security risks. Researchers have demonstrated how Notion's own agents could be tricked via prompt injection to exfiltrate private data using their built-in web search tool. This highlights a critical challenge for the industry: securing the AI "teammates" themselves, as their ability to autonomously interact with multiple services creates a new, complex attack surface.
