The release of Apple’s iOS 27, and with it the expansion of Siri AI functionalities as an agent rather than just a traditional assistant, marks a major shift in how we relate to our mobile devices. Until now, using an app meant manually finding it on your home screen, tapping to open it, and swiping through its menu. Over the past decade, we’ve also grown accustomed to AI assistants giving instructions and then doing the work for us.
With the new agentic AI features, Siri can now access the app, find what it can do, and trigger those actions without you having to manually handle your phone. That said, while the new technology offers a more seamless mobile experience, it also introduces new security concerns, especially given the access these AI agents need and the information some installed apps hold.
The shift to agentic workflows
Apple has been careful in how it describes these consumer agents, calling them exciting experiments. That changed recently, during their June 8th WWDC keynote, when they showcased a major overhaul to Siri’s workflow. They announced they will add an AI-assisted iOS 27 password-update workflow to the Password app.
The feature lets you open the Security tab, tap “Fix Passwords,” and then navigate flagged accounts and websites to change weak or compromised passwords. Apple’s App Intents framework makes this possible by letting app developers expose internal features and data to Apple Intelligence and Siri AI.
With App Intents, developers use native Swift code to control what their app does and the data it manages. The app’s defined actions are then automatically plugged into Siri for system-wide discoverability. App Intents act as Apple’s ecosystem’s core “action substrate” that Siri can use to trigger specific functions inside 3rd party apps.
However, this ability to retrieve information and trigger functions without the user opening the app raises another question about access. It means developers now have to be even more hands-on about what information and functions the assistant can reach. Surprisingly, a research study published by NowSecure last month found that 1 in 4 mobile apps it assessed over the last 3 months exposed functionality to Siri AI.
While the report didn’t establish whether these functionalities were exploited, it highlights another aspect of mobile application security. Traditionally, mobile security testing was not designed to evaluate AI-driven pathways. However, today, functions initially designed for harmless direct human interactions may now carry different risks when they interact with AI agents.
Online casino apps putting the automation to the test
The online casino app ecosystem best showcases the high-risk nature of autonomous agentic AIs. These apps run on high-risk components such as KYC identity verification, continuous real-money wagering, and linked credit cards, and are engineered for absolute financial precision. In Ontario, where the best online casino apps must comply with AGCO and iGaming Ontario rules requiring age checks, deposit limits, and self-exclusion tools, the stakes for accuracy get even higher.
Now drop an autonomous agent into that environment. For simple withdrawal-status retrieval, asking an AI assistant may be a sensible use of automation, since the margin for error is high. Now consider tasking the AI with initiating deposits or submitting withdrawal requests – higher-stakes capabilities. A slight misunderstanding of your instructions or indirect prompt injection from a malicious party could lead the AI to select the wrong option, with dire consequences.
Apple has acknowledged that they are conducting research on indirect prompt injection. These prompts could appear as hidden text in promotional emails or push notifications from malicious attackers and steer AI assistants that already have full authority over a user’s account. The same risk applies to online casinos, banking apps, crypto wallets, and trading platforms.
In Canada, where market regulatory requirements are strict, AI automation will need extensive testing to ensure it doesn’t remove the friction strategically placed as responsible-gambling safeguards.
Guardrails needed for AI agents before they act
The convenience of agentic workflows is undeniable. However, as iOS 27 and other software normalize autonomous app navigation, this convenience may cost us financial security. The new tech is a wake-up call for app developers, who now need to separate access to information from permission to change something.
This separation means an assistant you can prompt to read your transaction history shouldn’t also be able to automatically initiate a withdrawal from the account. For casino apps, this could mean building sustainable guardrails with dynamic authorization layers based on the function being executed. The best online casino apps will now be judged on more than just their welcome bonuses and game libraries. They’ll also be judged on how they design permissions for AI assistants. Ideally, placing a bet stays human, as does depositing; identity verification should also get the strictest treatment, while the AI assistant can read transaction histories and available game libraries.
As iOS 27 hits our mobile devices this Autumn, blanket accessibility is almost over. It’s wise to check every app that touches your money and review Siri’s permissions in each one.