Siri AI: safe for me, dangerous for thee

Apple did the tech-company-in-2026 thing and talked about AI at WWDC 2026. For roughly 40 minutes in a presentation that ran 72 minutes, it was essentially the only thing they talked about. I am sure it’s cool, or it’s another step toward the abyss. Your pick, honestly.

Something Apple forgot to mention while advertising the ability to annoy your friends, family, and coworkers with AI slop is that EU citizens won’t get access when iOS 27 releases. In fact, Apple currently has no timeline for when Siri AI — yes, that is what it is called now — will be available in the EU, according to their newsroom post.

Apple claims this is because of the European Commission’s “extreme interpretation of the DMA,” which would require Apple to grant third-party AI assistants “nearly unlimited access to a user’s device.” Now there is a scary thought. I sure wouldn’t want just any AI assistant to have that much access to my phone. I don’t have to make the argument for why that is an incredibly bad idea myself; Apple helpfully points out the dangers of the current crop of AI agents: “AI systems can be hijacked to steal personal data — like passwords and photos — and to permanently alter files and account settings without a user’s consent.”

And they don’t even have to be hijacked to do so. AI agents can also make bad judgment calls and mistakes all on their own. For your reading pleasure, in case you haven’t seen this story yet, here is Nick Davidov recounting his experience with Claude Cowork deleting 15 years’ worth of photos from his wife’s desktop. He had asked it to organize the desktop on his wife’s Mac, and that’s one way to do it, for sure. Notably, in this story, Apple was the hero. iCloud has a recovery feature that allows you to rescue deleted files for 30 days. Something to keep in the back of your mind when Apple rolls out Siri AI.

However, given the clear dangers to EU users and the regulators’ failure to acknowledge these risks, there is currently no timeline for Siri AI’s availability in the EU on iOS and iPadOS.

Is Apple a hero for standing up to the out-of-control regulators in Brussels too? Are they protecting their users from the dangers of untrustworthy AI systems like Claude mentioned above? Well, no. They are throwing what amounts to a corporate temper tantrum.

What Apple is fighting over with the EU is the Digital Markets Act (DMA). It requires gatekeepers to act less like, well, gatekeepers. It’s designed to prevent gatekeepers, like Apple for iOS and Google for Android, from designing systems that block competition with their own software and services. Specifically, in this case, the DMA requires Apple to enable other providers of AI assistants to access the same iOS functionality that Siri AI has. No preferential treatment for the first-party option.

Apple’s response is that third-party access is possible in principle, but that the kind of access AI assistants require is dangerous. And Apple is right about that part. There are real security and privacy risks in giving AI assistants system-wide access to a phone.

But Apple says they have a solution for those risks. They proposed the Trusted System Agent (TSA), “an intermediary that would allow virtual assistants to safely access the same features and capabilities as Siri AI for devices in the EU.” No further details were given, likely because the EU rejected this and every other proposal by Apple. Apple helpfully provides us with what they believe is the reason for this too: EU regulators just refuse to acknowledge the real dangers that their demands create.

Except, this story has a big hole in it. Why would the Commission reject a Siri AI rollout in the EU based on a security system? It’s not clear how Apple’s desire to protect the security and privacy of its users is touched by the DMA at all. The reason is that Siri AI is not routed through TSA, just third-parties. A security layer for AI assistants, just not when they are called Siri, is exactly the sort of thing that the DMA takes issue with.

Apple is more or less explicit about this: the TSA allows third-party virtual assistants in the EU to safely access the same capabilities as Siri. If what Apple is proposing is routing third-parties and Siri through the TSA they would be explicit about this. They would lament that the Commission is incoherent, not extreme, in their application of the DMA. In this very aggressive statement, Apple is silent exactly where it would help them the most to be loud.

Another tell is the timeline that Apple proposed: launch Siri AI with iOS 27 and then over the next 18 months develop the TSA. The Commission rejected this proposal, but notably Apple did not announce they have to hold back Siri AI in the EU until the entirety of the TSA is ready. They are explicit that they currently don’t have a timeline. So, it wasn’t the slow rollout that the Commission took issue with. The issue is that in the end Siri would still get special treatment.

All of this significantly weakens Apple’s positioning here. Are they claiming that Siri is special and doesn’t have any of the weaknesses and dangers of other AI systems? I haven’t seen them make this claim. It would be quite extraordinary if Apple had figured out AI alignment and just forgot to tell us about it.

A system that enforces permissions, allows them to be easily revoked, and requests explicit user consent for dangerous actions is what you want if you are going to have a system-wide AI agent with the capabilities that Apple advertised at WWDC. The Claude Cowork case has shown the dangers of letting an AI agent act without oversight. What’s missing from Apple’s side is why we wouldn’t want this for Siri too.

Apple’s aggressive communication around the DMA isn’t new. My favorite example is their 2021 white paper, “Building a Trusted Ecosystem for Millions of Apps.” It details all the scary consequences of being allowed to install software on your own device from developers who do not pay Apple rent — malware, in short. Apple admits in that paper that their on-device protections against malware can only go so far; users can still be tricked into granting permissions. To defend against that risk, Apple argues, a defense-in-depth approach is needed.

Now Apple seems to argue that Siri is safe enough and has no need for the extra security layer they proposed with TSA. They are angry that the DMA prevents them from releasing Siri AI in the EU on their own terms. If there is a security and privacy concern here — and there is — Siri should have to adhere to the same security standards as third-party AI assistants. That also aligns with what the DMA requires in this case: evenhandedness.