prompt injection
973 MCP Packages, 71% Single-Maintainer: A Practitioner’s Guide to AI Developer Security
At a Glance AI security tooling adoption lags behind AI coding tool adoption by an order of magnitude. Download ratios: 10:1 on PyPI, 28:1 on npm. AI-generated code ships vulnerable at baseline. 45% failure ...
Access Controls Validate Identity. They Cannot Validate Intent. With AI Agents, That Distinction Is Now Critical
Access controls can confirm who or what is allowed to act. They cannot always tell whether the action makes sense. That gap becomes dangerous with AI agents, which can call tools, trigger ...
Guarding AI Agents: Boundaries and Safeguards
AI agents are useful, but they become risky when they can take action in real systems. In this episode, Tom Eston discusses recent reporting about attackers tricking Meta’s AI support chatbot into ...
When the Frontier Blinks: What the Mythos and Fable Controversy Reveals About AI Security
When Anthropic abruptly pulled Mythos 5 and Fable 5 from circulation, the move sent a jolt through the AI and cybersecurity communities. These were not minor point releases. They were widely regarded ...
The Half of Agent Security You’re Not Governing
The governance of AI agents faces a fundamental asymmetry: while MCP servers provide structured logs, the "Skills" that drive agent reasoning remain forensic black holes. As high-risk capabilities—such as arbitrary code execution ...
When AI Billing Breaks Trust: What the Claude Code Backlash Says About AI Governance
When AI Billing Breaks Trust: Lessons from the Claude Code Backlash AI adoption is accelerating, but trust is still fragile ...
Capsule Security Emerges From Stealth to Secure AI Agents at Runtime
Capsule Security emerges from stealth with a $7M seed round to launch a runtime security platform for AI agents. Featuring the open-source ClawGuard, the platform enforces governance and mitigates prompt injection risks ...
Bypassing LLM Supervisor Agents Through Indirect Prompt Injection
Indirect prompt injection lets attackers bypass LLM supervisor agents by hiding malicious instructions in profile fields and contextual data. Learn how this attack works and how to defend against it. The post ...
The Arms Race is Already Over. You Just Don’t Know Which Side Won.Â
Anthropic’s Claude 4.6 found 500+ zero-days, but the real story is economic. As AI secures code, attackers are shifting to the "Trust Layer"—AI-driven social engineering and identity deception ...
Which Came First: The System Prompt, or the RCE?
During a recent penetration test, we came across an AI-powered desktop application that acted as a bridge between Claude (Opus 4.5) and a third-party asset management platform. The idea is simple: instead ...
