How Indirect Prompt Injection Exploits RAG Pipelines And 4 Controls That Actually Contain It
When your LLM retrieves documents, emails, or web pages to answer queries, every one of those sources is a potential injection vector. Here is how indirect prompt injection works inside RAG architectures and what technical controls reduce your exposure.
Shopify Turns AI Agents Into Store Operators With Its New AI Toolkit
Shopify has launched an AI Toolkit that lets agents such as Claude Code, Cursor, and VS Code connect directly to store operations, pushing agentic commerce closer to real merchant workflows.
OpenAI Adds a $100 ChatGPT Pro Tier for Heavy Codex Users
OpenAI has introduced a new $100 monthly ChatGPT Pro plan aimed at people who use Codex heavily, creating a middle tier between Plus and the company's highest-priced offering.
Meta Launches Muse Spark as Its First Superintelligence Model and a Reset for Meta AI
Meta has introduced Muse Spark, the first model from Meta Superintelligence Labs, pairing multimodal reasoning, tool use, and multi-agent orchestration with a broader push to rebuild Meta AI around what Mark Zuckerberg calls personal superintelligence.
Amazon Rebuilds Kindle Scribe Around AI Notes and Its First Color Writing Experience
Amazon has unveiled a redesigned Kindle Scribe lineup that adds AI-powered notebook search, cloud document integrations, and its first color Scribe, turning the device into a clearer bet on productivity-focused AI hardware.
Uber Deepens AWS Partnership to Power Real-Time AI at Global Scale
Uber is expanding its use of AWS custom chips, pairing Graviton4 for millisecond-level operations with a Trainium3 pilot aimed at training larger, smarter AI models across its ride and delivery network.
Perplexity's Pivot Into AI Agents Sends Revenue Up 50%
Perplexity's push beyond AI search and into agentic tools is translating into faster commercial growth. The company reportedly crossed $450 million in annual recurring revenue in March after a 50% jump in revenue over the last month.
5 Types of Prompt Injection Attacks Targeting Deployed LLMs And How to Block Each One
Not all prompt injection attacks work the same way. This breakdown covers direct injection, indirect injection, jailbreaks, role-playing exploits, and multi-turn manipulation, with concrete defense controls for each attack type.
Anthropic Unveils Project Glasswing to Put Frontier AI on Cyber Defense Duty
Anthropic has launched Project Glasswing, a new initiative built around Claude Mythos Preview to help secure critical software before advanced AI systems make cyberattacks easier to scale. The company is framing it as a defense-first response to rapidly improving AI vulnerability research.
Open-Source Claude Code Job Search System Goes Viral After Helping Its Creator Get Hired
A builder has open-sourced Career-Ops, a Claude Code-powered job search system that evaluated 700-plus roles, generated tailored resumes, and reportedly helped land a Head of Applied AI job.
Microsoft Says Copilot's 'Entertainment Only' Warning Is Legacy Language
Microsoft is facing fresh scrutiny over Copilot's terms of use after users resurfaced language calling the product "for entertainment purposes only," even as the company pushes businesses to adopt it more deeply.
Anthropic Moves OpenClaw Usage Outside Claude Code Subscriptions
Anthropic is separating OpenClaw and other third-party harness usage from Claude Code subscriptions, forcing affected developers onto pay-as-you-go billing and adding fresh tension to the fast-growing AI coding tools market.