What happened
An OSS developer building a2a-rails and ActingFor argues AI agents that send email, buy goods, delete files or update databases turn mistakes into real-world actions, making limits on what agents may do as important as capability.
Why it matters
Because an agent's errors now hit real systems, money and data, the developer says permission controls set outside the AI are needed to decide which actions a given user's agent may perform.
WHO IT HITSDevelopers and platform teams wiring AI agents into payment, email or database tools will need per-action permission checks that sit outside the model, rather than trusting the model's own judgment.
Summaries like this, in your inbox every morning.
The developer's argument rests on a shift already visible in the tooling around AI agents. Protocols such as MCP, which lets AI use external tools and data, and A2A, which lets separate AI agents hand work to each other, make it easier for agents to reach outside systems. The developer's own projects sit right at that seam: a2a-rails exposes a Rails application as an A2A-capable agent, while ActingFor handles proxy authorization. Notably, the second gem is framed as a check outside the AI rather than a stronger model inside it.
Institutional work has moved in parallel. In February 2026, NIST announced an AI Agent Standards Initiative that covers not only interoperability but also security, identity and authorization. OWASP's Top 10 for Agentic Applications 2026 catalogues risks such as attackers tampering with an agent's goals, tool misuse and improper use of privileges. The developer reads this as evidence that agent safety is a concrete engineering problem rather than a distant worry, and expects corporate adoption of agents to raise safety requirements even as building agents gets easier.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Chunghwa Telecom president Rong-shy Lin said the shift from conversational AI to autonomous AI agents is resha…
Apple and Meta said they would jointly release a new open-weight AI model designed to run on high-end Macs, ai…
A writer and entrepreneur opened his daily to-do list and found 156 items, and concluded that an AI assistant…
A Zenn contributor sorted AI use into three modes — AI as an assistant, AI running alongside people, and AI th…
A non-engineer operator says one person ships four smartphone apps on Google Play, three on the App Store, wit…
A developer released agent-guardrails, a one-script Python standard-library plugin that guards Bash, Write, Ed…
