NEW YORK, Nov 20 (Reuters) - Encrypted messaging service Signal said on Friday it would restrict screenshots to stop AI agents from reading private chats on phones, warning that the tech industry's push to embed AI in consumer devices was eroding its ability to guarantee privacy.
Signal, widely regarded by privacy advocates as a leader in secure messaging, called its new protections "incomplete" and urged operating system providers Apple, Alphabet's Google and Microsoft to give app developers more concrete ways to block the agents from accessing sensitive data.
It also called on those companies to provide more information about how their agents function, so developers can assess what data the bots are accessing and adjust their systems accordingly.
The move reflects growing tension between privacy-focused apps and the biggest tech companies, which are racing to roll out AI agents designed to see what is on a user's screen and take actions based on it.
Signal identified as particularly worrisome the AI capabilities built directly into operating systems, like Apple's Siri AI and Google's Gemini Intelligence, because of how deeply they are integrated into people's devices.
Operating systems maintain the environment in which Signal and other applications must exist,the company said.
"Right now, these agentic AI integrations are changing this environment in a way that’s making it increasingly hostile to privacy. This is an existential threat to Signal, and our ability to responsibly uphold our mission," it said in a blog post provided to Reuters ahead of its release on Signal's website.
To prevent agents from scanning Signal messages, the service is introducing new settings blocking screenshot access on people's phones and switching those restrictions on by default. Users will receive pop-up windows giving them options to disable the blocks when Signal detects screenshot attempts or efforts by an AI assistant to access the screen, depending on the device.
Signal similarly blocked Microsoft's Recall AI from taking screenshots on Windows last year. It said it was unable to reliably offer suchprotections on Apple's macOS.
GETTING AROUND PROTECTIONS
Agents that can take actions on behalf of users are billed as the next big wave of AI technology, capable of surfacing relevant information from across different apps and executing time-saving tasks like answering emails, booking appointments and making online purchases autonomously, without direct instructions.
However, that reach can get around the protections that apps like Signal build for themselves.
Reuters reported earlier this week that Meta's Muse agent was able to bypass its own official "Connector" tool toaccess Signal messages on both an Apple computer and a phone using Google's Android system.In both cases, Muse used broad operating system-level permissions granted during the app's initial setup to pull in the messages, using those authorizations in ways a user might not anticipate.
On Apple's macOS, for instance, it used a combination of the computer’s accessibility and screenshot tools,repurposing settings Apple offers for people with disabilities to navigate over to Signal, take screenshots of chats there, extract text from those pictures and reproduce the messages inside the Muse app.
An operating system update from Apple this week effectively closed that specific pathway, but also introduced a new set of permissions aimed at agents that offers similar capabilities paired with additional user warnings.
Meta said in response to Reuters' reporting that agents need to see and interact with people’s computers in order to work on their behalf, and that Muse can access other apps only via permissions users explicitly grant.
Signal said that developers should have straightforward ways of opting their apps out of such data processing by AI assistants entirely, specifically mentioning screen reading, notifications and other file system access.
The service, owned by a non-profit foundation, promises its users a secure messaging experience by limiting the data it stores on its own servers, eschewing ads and offering end-to-end encryption, meaning it cannot access and read the conversations conducted on its app.
AI systems that scan people's screens can undermine those protections by pulling the messages out of Signal's secure environment and into whatever other servers the agent is using, it said in the post.
(Reporting by Katie Paul in New York; Editing by Sonali Paul)
