Blog

Promptfoo is the most widely adopted open-source AI red teaming tool, with 18K+ GitHub stars and use at 25% of Fortune 500 companies. OpenAI acquired it in March 2026 for approximately $86M. This post covers how Promptfoo’s automated red teaming works, the 50+ vulnerability types it catches, how it compares to alternatives like PyRIT and Garak, and what the acquisition signals about the future of AI agent security testing.

Forrester’s 2026 predictions declare the AI hype period over. Only 15% of AI decision-makers report EBITDA improvement. Gartner expects 40% of agentic AI projects canceled by 2027. RAND puts the broader failure rate at 80%. This post breaks down what the 15% who see returns actually do differently.

PayPal, Sabre, and Mindtrip announced the travel industry’s first complete agentic AI experience: one flow that handles trip discovery, booking across 420+ airlines and 750,000+ hotels, and payment. Launching Q2 2026 inside the PayPal app, it is the first time all three layers run through AI agents without handing off to external sites.

Anthropic released Agent Skills as an open standard on December 18, 2025. Within 48 hours, Microsoft and OpenAI adopted the identical format. Three months later, 32 tools from competing companies read the same SKILL.md files. 100K GitHub stars. 89,000+ skills on marketplaces. This is the fastest cross-vendor standardization event in AI tooling history, and it happened because the spec fits in a single README.

Gartner estimates only about 130 of thousands of vendors claiming agentic AI capabilities actually deliver autonomous, goal-pursuing systems. The rest are rebranding chatbots, RPA bots, and workflow automation as agents. With 40% of agentic AI projects headed for cancellation by 2027, this post breaks down the five traits that separate real agents from rebranded tools, and provides a concrete vendor evaluation checklist.

Every AI coding agent forgets everything the moment you close your terminal. claude-mem fixes that with five lifecycle hooks, a dual SQLite/Chroma storage layer, and AI-powered compression that achieves 10x token efficiency. The plugin rocketed to #1 on GitHub Trending and now has over 21,000 stars. Here is how it works and why memory infrastructure is becoming the critical layer for agentic coding.

Greenhouse’s 2025 report reveals the widest trust gap in modern recruiting: hiring managers believe in AI, candidates do not. Both sides now actively game AI systems, from white-font keyword stuffing to prompt injection on resumes. This post examines the specific cheating tactics, the Greenhouse data, and the lawsuits that are forcing a reckoning.

Zenity contributed 14 new agentic AI attack techniques to MITRE ATLAS in early 2026, covering Agent Context Poisoning, Exfiltration via Tool Invocation, Memory Manipulation, and more. Combined with the OpenClaw investigation that documented 7 agent-specific techniques from real incidents, ATLAS is now the authoritative threat taxonomy for AI agents. This post breaks down every new technique, compares ATLAS to OWASP’s agentic Top 10, and provides a concrete detection roadmap for SOC teams.

AI tools promise efficiency but deliver escalation. UC Berkeley researchers Aruna Ranganathan and Hatim Rahman found that AI creates a workplace Jevons Paradox: faster task completion leads to more tasks, higher standards, and burnout. 77% of AI-using employees say their workload increased. This post breaks down the research and offers concrete strategies to escape the trap.

Oracle released 16 purpose-built AI agents across marketing, sales, service, and field service inside Fusion Cloud CX, included at no additional cost. The pricing move directly undercuts Salesforce Agentforce’s $2-per-conversation model. Here is what the agents actually do, how they compare to the competition, and what the unified data layer means for enterprise CX stacks.