Anthropic has introduced Claude for Chrome, a browser extension that embeds its Claude AI directly into web pages. The tool can summarize content, analyze products, manage calendars, and even test website features. Launched as a pilot for 1,000 Max‑plan subscribers at $200 per month, the extension includes site‑level permissions and prompts before high‑risk actions to address prompt‑injection vulnerabilities. Safety updates reduced the success rate of malicious instructions from 23.6% to with safeguards that block high‑risk categories. Claude for Chrome joins a growing roster of AI‑powered browsers, including Perplexity’s Comet, Google’s Gemini, Microsoft Edge Copilot, Brave’s Leo, and Opera’s Aria. Read more →

Anthropic’s Claude AI Chrome extension was tested against dozens of attack scenarios, showing a significant success rate when safety mitigations were absent. The company introduced site‑level permissions and user confirmations that cut the success rate, but independent researchers still label the remaining risk as catastrophic. Recent related incidents involving Perplexity’s Comet extension underscore the broader challenge of securing AI agents in browsers. Anthropic plans further testing before broader release, emphasizing the need for stronger protections and user awareness. Read more →

Google is introducing an AI-powered age-estimation system to protect minors from harmful content, while the UK's age-verification law has been met with VPN downloads and criticism over security and privacy risks. Read more →

Enterprises are abandoning Retrieval-Augmented Generation (RAG) architectures in favour of agent-based approaches due to security and performance limitations. Agent-based architectures offer several advantages, including elimination of duplicate data repositories and preservation of authorization models. Read more →