Developer releases AI Guard Gateway v0.1.0 to block prompt injection and endpoint hijacking

·3 views

A developer known as MagoPredator has released AI Guard Gateway v0.1.0, an open-source reverse proxy designed to secure AI inference APIs such as Ollama and LiteLLM against unauthorized access and abuse. The tool was built in response to documented vulnerabilities involving exposed AI endpoints that can be exploited without any traditional exploit, simply by accessing unsecured APIs anonymously. AI Guard Gateway addresses key threat vectors including endpoint hijacking, prompt injection, resource exhaustion, and PII leakage through components such as mandatory API key or JWT authentication, sliding-window rate limiting, and automatic redaction of sensitive data. The project was developed using Spec-Driven Development and validated through pytest test coverage, a Bandit security scan returning zero high or medium severity findings, and continuous analysis via a SonarCloud CI/CD pipeline. The gateway is available on GitHub under the AGPL-3.0 license and is intended to be deployed in front of any language model serving infrastructure.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Historical facts that reframe your sense of time, from Oxford to the Aztecs

A Smithsonian Magazine article highlights surprising historical comparisons that challenge common assumptions about the timeline of human civilization. One notable example is that the University of Oxford predates the founding of the Aztec Empire, illustrating how institutions can be older than entire civilizations. The piece presents several such juxtapositions to help readers gain a fresh perspective on how events in history overlap in unexpected ways. The article was shared on Hacker News, where it attracted modest attention with five points and no comments at the time of posting.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

Osloq AI Agent Reproduces Bugs and Reports Root Causes Without Touching Your Code

Osloq is an AI agent designed to investigate software bugs by reading GitHub issues, tracing code paths, and reproducing problems inside an isolated sandbox environment. Unlike tools such as Devin or Sweep AI that automatically write fixes and open pull requests, Osloq focuses solely on identifying and documenting the root cause, leaving all fix decisions to the developer. Once a bug is reproduced, Osloq generates a detailed report containing screenshots, console logs, call stacks, and a plain-language explanation of what went wrong. The sandbox is destroyed after each investigation, and the tool operates with read-only access to repositories, meaning no code is stored or used for model training. Osloq positions itself as a low-risk "investigator" tool suited for QA teams, open-source maintainers, and safety-critical projects where evidence-based decision-making is a priority.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Security Scan Finds 332 Critical Flaws Across LlamaIndex, AutoGen, and CrewAI

A security audit using AgentGuard v0.6.1 uncovered 332 critical vulnerabilities across three widely used AI agent frameworks: LlamaIndex, AutoGen, and CrewAI. LlamaIndex alone accounted for 252 critical findings, including credential exposure in replay logs and unsafe trust boundary handling in its MCP host. CrewAI showed 391 medium-severity findings, with agent data flowing to external endpoints without proper constraints. All three frameworks are in active production use, with some boasting over 30,000 GitHub stars and deployments at Fortune 500 companies. The researchers note that fixes exist for all identified issues, including input validation, sandbox enforcement, and log scrubbing, representing standard application security practices not yet consistently applied to AI agent code.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

CPMO Playbook Chapter 9: How Product Leaders Should Navigate the Scale Stage

Chapter 9 of 'The CPMO Playbook' by Ali Sadhik Shaik focuses on what scaling means for a Chief Product and Marketing Officer. The chapter outlines team structures at scale, covering roles such as PM, PMM, Growth, Brand, and Ops. It details key operational cadences including weekly reviews, monthly business reviews, and quarterly planning cycles. Core metrics discussed include Net Revenue Retention, segment win rate, and pipeline coverage. The chapter also warns against a common trap: over-optimizing the funnel while the broader product category is shifting.

0 comments Read more at DEV Community