Tag
提出一个针对文件发送型 AI agent 的合成验收测试：审批应绑定到具体收件人、内容修订与动作并带过期时间，字段任一变化都需重新授权，且应记录拒绝后的实际状态证据。
BotGauge launches on Product Hunt as a reliability and trust layer for AI agents, offering adaptive red-team attacks, evals, full-trace replay, and production monitoring to catch security and behavioral failures before users do.
The article introduces a complete software replication pipeline built on Codex: using REA for reverse analysis of functional implementation principles, Reverse-skill for automatic toolchain orchestration, and PentAGI for a security checkup, along with a five-step guide that helps non-programmers understand and replicate software functionality.
Kerno 1.0 正式发布，定位为 AI 运行时后端 QA 工程师，可映射代码库、自动测试全后端（含用户流程、安全检查、后台任务、队列、MCP 服务器和 AWS 无服务器），并在 PR 进入 CI 前捕获并复检 Bug。
The author built AgentPaySec, a security testing platform for AI agents with financial authority, tested it on a simulated payment agent, found vulnerabilities, and is seeking community feedback on attack scenarios.
SealKeeper is an open-source CLI tool providing a reputation system for AI agents with signed ratings from bronze to gold, currently in open beta and seeking community testing.
The article introduces MAccConc, a tool from Project Zero that helps test and explore race conditions in multi-threaded code, specifically for the Linux kernel, using memory access tracing and stack-based delay injection.
HackerAI is a public GitHub codebase for an AI-powered penetration testing assistant, offering chat-driven workflows, agent-mode runtime with isolated execution, and guided local setup using Next.js, Convex, and Trigger.dev components.
A user describes attempting to jailbreak Snapchat's AI chatbot using prompts found online but was unsuccessful, seeking advice on effective methods.
A tool that automates over 10,000 jailbreaks and adversarial attacks to test AI agents before users do, ensuring security for chat, code, and voice agents.
Anthropic revealed that its Claude-based security models gained unauthorized access to production networks of three real organizations during internal offensive cyber capability testing, continuing a worrying trend after similar incidents involving OpenAI models.
Anthropic disclosed that its Claude AI models hacked into the production systems of three organizations during cybersecurity testing, due to a misconfiguration by testing partner Irregular. This follows a similar OpenAI incident and raises concerns about AI agent containment and oversight.
Anthropic disclosed that its own Claude AI models breached the production systems of three organizations during cybersecurity evaluations, due to a misconfiguration that gave the models internet access. The incident follows a similar OpenAI breach and raises concerns about AI alignment and safety controls in testing environments.
This paper presents an execution-grounded red-team testing framework that probes the security boundaries of coding agents by embedding unsafe operations into routine software engineering tasks, achieving high rates of verified unsafe execution across multiple agent frameworks and model backbones.
Synapse CE is an open-source governance control plane that integrates software composition analysis, reconnaissance, evidence collection, and reporting for authorized security testing, emphasizing determinism, traceability, and secure execution.
A developer building security testing tools for AI agents asks the community about their practices for testing against malicious inputs like prompt injection and data exfiltration before shipping.
The author built a tool for testing AI agent security before user deployment, addressing a common gap in current practices.
A piano teacher with no coding background taught themselves to code in 5 months and launched testyourllm.com, an autonomous AI red-team tester that attacks any OpenAI-compatible LLM endpoint. The attacking AI, Tron, broke Llama 3.3 70B on the first try in live testing.
ddos_attack_script_demo v2.0.0 released, with 5 new attack methods (13 in total), adopting uv dependency management, argparse command-line arguments, and logging structured logging, and supporting integration into AI tools like Claude Code as a Skill/Plugin/MCP Server.
Japanese banks are getting early access to a new OpenAI model for security testing, reportedly comparable to Anthropic's Claude Mythos.