Apple Writes the Rules for Mac Agents While Hiring to Build Its Own
Bloomberg's report rests on a job listing; Apple hasn't confirmed anything, and no source shows the macOS 27.3 warnings are tied to the agent project.

Apple hasn’t confirmed a computer-use AI agent, yet Bloomberg spotted a job listing for a general-purpose model targeted for late 2027. That comes as the macOS 27.3 beta adds strict warning prompts that slow down rival tools like ChatGPT.
Apple calls that protection. Some AI companies and users, per Bloomberg’s report, see a brake on competitors. The sources can’t settle which, and shouldn’t be asked to.
What they show is a company writing the permission rules for Mac agents while recruiting someone to build its own.
What the listing actually says
TechCrunch says Bloomberg found a listing for a senior engineering manager over an “AI agent that can use a computer.”
It would browse the web, work across apps, run commands, write and debug code, and call services, choosing the right tool at each step.
Gizmodo adds that it would coordinate specialized subagents and work “privately,” and calls the listing’s language “more of a wish list” than a features list.
The listing also says agents that run code and act for a user need “strong permission models, sandboxing, and transparency,” per Gizmodo.
Neither outlet says whether the agent is only for Mac devices. TechCrunch frames it around the Mac, while Gizmodo notes the listing mentions bridging devices. Name, price, Siri link, and final product decision are all unstated.
What’s happening to everyone else’s agents
The macOS 27.3 beta shows warnings when users grant third-party agents control of a Mac, and ChatGPT’s computer-use setup now takes 14 steps instead of four, per Gizmodo citing Bloomberg.
Google’s Gemini gained Mac control earlier the same Thursday, while Meta’s recently launched Muse Mac agent has already drawn scrutiny, underscoring why Apple is tightening access rules across third-party tools.
Neither source says if the 14 steps hit other agents, or if the final release keeps them.
What to make of the timing
The dates line up, and that’s all the sources establish. Apple’s published rules cover backup and accessibility apps as well, so calling them an anti-agent measure would go past the evidence.
Calling them neutral would too. A platform owner that sets permission prompts and plans a competing product has a conflict of interest, whatever its intent, and a fair reader keeps both facts in view.
For a company running agents on Macs, the practical points are modest. Expect more setup friction in 27.3 if the beta holds.
Don’t wait for Apple’s agent: a launch as soon as the end of 2027 is a hope, not a schedule. And the listing’s own sentence makes a decent checklist for any agent, Apple’s included: permissions, sandboxing, transparency.
Source: Apple Working on Own AI Agent to Rival OpenAI and Anthropic



