Brief IA
›

Signal limits screenshots against AI agents on iOS 27

AI Tools·Tom Levy·

Signal limits screenshots against AI agents on iOS 27

Signal limits screenshots against AI agents on iOS 27
Key Takeaways
1Signal plans to block screenshots on iOS 27 and those attempted by AI assistants on Android.
2According to Signal, its protections currently do not work on macOS.
3A setting will allow users to disable the restriction, an option strongly discouraged by the messaging app.
Why it matters : according to Meredith Whittaker, agents can also access data through other means, including the file system.

Signal plans to block screenshots on iOS 27 and those taken by AI assistants on Android, announced its president, Meredith Whittaker, on Friday. The messaging platform cites the privacy of exchanges in the face of agents integrated into operating systems but acknowledges that these restrictions do not cover all possible access points.

Incomplete Protections Against Agents

On macOS, the protections are currently ineffective, Signal indicates. Meredith Whittaker also acknowledges that blocking screenshots is not enough: an agent can, according to her, retrieve content stored on the device by accessing the file system.

The permissions granted to Muse, Meta's agent, during its initial setup are being used in unexpected ways. Thanks to these permissions, Muse accessed applications and data that the user had not explicitly linked to it.

For the president of Signal, developers do not have sufficiently robust tools to protect the privacy of their applications against these agents, which she regrets. According to her, even when access can be detected and then blocked, Google, Apple, Microsoft, and others do not allow it to be restricted easily without severely degrading basic functionalities.

Users Will Be Able to Lift the Block

A setting will allow users to restore screenshots. Signal strongly advises against this deactivation: Meredith Whittaker warns that it could expose the user's content as well as that of their contacts to agents, undermining privacy guarantees.

La lettre du soirGratuit · 19 h

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

The messaging platform had already implemented a block last year against Recall AI from Microsoft. This feature could capture images of Signal conversations on devices running Windows 11.

Google DeepMind Checks Signal's Code

Last week, researchers from Google DeepMind used AI agents to conduct a formal verification of Signal's code. They relied on the messaging platform's work aimed at demonstrating the proper functioning of encryption. The concerns expressed by Meredith Whittaker focus on another access point: reading messages once they are displayed.

She specifically points to Gemini Intelligence from Google and Siri AI's Onscreen Awareness from Apple, which observe the screen to respond to requests. According to her, sensitive data can be captured, stored, and then accessed by local databases or remote servers used by the agent. Signal then loses control over its fate, which threatens, she warns, the privacy of every participant in the exchanges.

These functions are integrated into the operating systems, while OpenAI and Anthropic's chatbots are external services to these systems. Samsung is also reportedly working on an agent called Galaxy Claw, while Apple is developing a competitor to Claude and ChatGPT.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.