Full Summary
This Sunday morning, Meta launches Muse Code, its first dedicated AI coding agent, now in beta for macOS and Linux. Both Memeburn and t.co confirm this terminal-based tool uses Muse Spark 1.2 to write, debug, and verify code, aiming for complete software-engineering jobs, not just short snippets. It can plan changes, edit multiple files, and test software across large code repositories. What's interesting is its persistent activity log, enabling crash recovery, and Meta's strategy to price it well below cost—twenty cents for a million output tokens—in exchange for developer prompts and completions as training data. This directly challenges existing platforms like Anthropic’s Claude Code and OpenAI’s Codex, with Meta competing on crash safety and data acquisition. Meanwhile, Agentic AI is making waves across various sectors. The UAE is embarking on a national project, confirmed by ARN News Centre, Dubai Eye 103.8, The Tribune, and Asianet Newsable, to convert 50% of federal government operations to agentic AI within two years. More than 100 officials attended a workshop for this initiative, guided by the principle "human leads, AI enables." This aims to enhance government efficiency and decision-making in policy development and future planning. In cybersecurity, BankInfoSecurity reports that agentic AI is already transforming Security Operations Centers, or SOCs, by integrating into every stage of threat workflow, from detection to response. This boosts efficiency and allows for deeper investigations, with experts from Optiv and Google Cloud emphasizing AI's role in enhancing, not replacing, human decision-making. However, a new security risk emerges: "Agentjacking." Forkast.news reveals this attack vector, which exploits publicly exposed Sentry Data Source Names and AI coding agents, allowing remote code execution on a developer's machine. Controlled tests showed an 85% success rate across over 100 organizations, with attackers able to steal credentials like AWS keys and GitHub tokens. Tenet has released "agent-jackstop" to harden configurations for Cursor and Claude Code, blocking malicious package fetches. This vulnerability highlights significant risks for developers using AI coding agents. This shift towards agentic AI means new opportunities and new risks are rapidly emerging, impacting how businesses operate and how governments serve their citizens.