Signal to restrict screenshots over AI agent concerns – again

1 day ago 9

ZDNET’s cardinal takeaways

  • Secure messenger Signal is disabling screenshots.
  • The determination is successful effect to an uptick successful AI agents from large OS developers.
  • Agents tin undermine Signal’s ngo to support chats private.

Privacy concerns astir AI agents person prompted encrypted messaging app Signal to acceptable caller limits connected screenshots. The institution volition forestall screenshotting successful its iOS 27 app and “when AI assistants effort to seizure surface information connected Android,” Signal President Meredith Whittaker wrote successful a blog post Friday.

Whittaker noted that Google and Apple’s agentic features, including Gemini Intelligence and Siri AI’s Onscreen Awareness, compromise the privateness missions of third-party developers. These features observe what’s connected a user’s surface arsenic discourse for completing requests.

Also: My 5 favourite Siri AI ‘onscreen awareness’ tricks – and why

“Your delicate information is being captured, stored, and made accessible to immoderate section database oregon distant server an OS-level AI cause is utilizing astatine that moment,” Whittaker said. “All of this is extracurricular of Signal’s power erstwhile the information has been captured, and imperils the privateness of each the participants successful a conversation.”

Signal besides blocked screenshots past twelvemonth successful effect to Microsoft’s controversial Recall AI feature, which could seizure screenshots of Signal chats connected devices moving Windows 11.

More agents, much risks

Signal’s determination is different reminder of however Big Tech’s AI strategy trickles downstream. As AI features person moved from services outer to the OS (like chatbots from OpenAI oregon Anthropic) to further wrong of it, third-party apps person to accommodate to conditions that undermine their hard-won privateness infrastructure.

Also: How I usage Meta Muse safely to agelong my AI fund – and what I ne'er fto it do

Just past week, researchers astatine Google DeepMind utilized AI agents to formally verify Signal’s code, gathering connected Signal’s ain enactment to fundamentally beryllium the app’s encryption works. That’s a precocious barroom for a information warrant wrong the app, but it becomes meaningless if an cause tin simply drawback messages disconnected a screen.

Samsung is reportedly moving connected its ain agent, Galaxy Claw, and Apple is developing a rival for Claude and ChatGPT arsenic well.

Whittaker noted Signal’s screenshot regularisation isn’t perfect, saying that AI agents tin inactive entree app information done different means, “such arsenic retrieving contented already contiguous connected a user’s instrumentality via the filesystem.” Earlier this week, a Reuters investigation supported by Signal and the AI Now Institute recovered that Meta’s Muse AI cause was utilizing permissions from its archetypal setup successful unexpected ways to entree apps and information that a idiosyncratic had not explicitly connected it to.

Signal said its protections don’t presently enactment connected MacOS.

“It is simply a superior occupation that Signal – and exertion developers beyond – deficiency robust tools to forestall AI agents from undermining application-level privacy,” she added. “Even wherever we are capable to observe and artifact AI information access, Google, Apple, Microsoft, and others bash not supply ways for developers to easy restrict specified AI information entree without making drastic compromises to the halfway idiosyncratic experience.”

How to disable the restriction

If you’d inactive similar to instrumentality screenshots, you tin simply disable the setting, though Signal’s announcement powerfully advised against it.

“Doing truthful volition perchance exposure your (and your contacts’) Signal contented to these AI agents, and interruption our privateness guarantees,” Whittaker said.

Radhika Rajkumar

Radhika Rajkumar

Senior Editor

Radhika Rajkumar is simply a elder exertion astatine ZDNET based successful New York City. She covers AI, specializing successful safety, privateness and security, policy, education, and synthetic media. She besides leads ZDNET's newsletter strategy. Radhika holds a Masters successful Creative Publishing and Critical Journalism from The New School.

Read Entire Article