AI Governance Institute
← News

Signal Blocks OS AI Agents, But the Data Leakage Gap Remains on macOS

What happened

Signal updated its iOS and Android applications to block operating system-integrated AI agents from capturing screenshots of private messages by default. Blocked agents include Apple Siri, Google Gemini, and other on-device agents. In a blog post, CEO Meredith Whittaker explained that these agents scan on-screen content and may transmit it to cloud servers, undermining the application's end-to-end encryption. Whittaker acknowledged that screenshot blocking is only a partial fix, because agents may still access application data through the device file system by other means. She directly criticized Apple, Google, and Microsoft for not giving developers reliable tools to restrict AI agent data access without degrading the core user experience. macOS protections are not yet possible. Apple has not provided the platform controls needed. A macOS update with more granular data access permissions is in public beta as macOS 27.3 but is not yet available. Users who disable screenshot protection in settings risk exposing their own messages and those of their contacts to AI agents, as reported alongside broader concerns about OS-layer agent access.

Why it matters

  • ·Employees using Signal on personal or corporate devices for sensitive communications face a real risk. OS-integrated AI agents may read that content and transmit it to cloud servers. This bypasses enterprise data classification and confidentiality controls entirely. This creates unaddressed exposure under regulations such as the General Data Protection Regulation (GDPR), particularly where messages cross borders or involve personal data.
  • ·The gap is structural, not incidental: Signal's CEO confirmed that platform owners do not give developers the tools needed to block AI agent access reliably. This means compliance teams cannot resolve the risk through application settings or mobile device management policies alone; the exposure depends on OS vendor decisions outside the enterprise's control.
  • ·macOS devices remain fully unprotected by Signal's new controls, and the platform fix is in beta testing only. Organizations where employees use macOS for communications, including legal, HR, finance, or executive functions, carry a live and unmitigated risk of OS-layer AI agents reading sensitive on-screen content.

Stay ahead of stories like this

Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Free every Thursday. Unsubscribe anytime.

Governance controls affected

What to do now

  • ☐Determine whether employees use Signal or similarly privacy-sensitive messaging apps on corporate-managed iOS, Android, or macOS devices, and whether OS-integrated AI agents such as Siri or Gemini are enabled on those devices.
  • ☐Review mobile device management policies to assess whether they can disable or restrict OS-level AI agent features (such as Siri's on-screen awareness or Gemini's context-scanning) on managed endpoints, and implement restrictions where possible.
  • ☐Issue an interim advisory to employees who use macOS for sensitive communications, noting that Signal's screenshot protections are not yet available on macOS and that OS-level AI agents may be able to read on-screen content.
  • ☐Assess whether sensitive communications policies need to be updated to address the risk that OS-integrated AI agents may transmit on-screen content to third-party cloud servers, including cross-border transfer implications under applicable data protection obligations.
  • ☐Add OS-layer AI agent access controls to the next vendor and platform review cycle for Apple, Google, and Microsoft, tracking progress on macOS 27.3 platform controls and any equivalent Android or iOS updates.

What to watch next

Compliance teams should monitor Apple's rollout of macOS 27.3. It is expected to introduce more granular data access controls. These could allow developers like Signal to restrict AI agent access on desktop. Progress on the Five Eyes Guidance on the Careful Adoption of Agentic AI Services may also produce updated device-level recommendations as OS-integrated agents proliferate. Track broader regulatory signals around AI agent data access at the OS layer. This includes guidance from data protection authorities under the General Data Protection Regulation (GDPR) or national equivalents. This structural gap is becoming more visible to regulators. The pattern of major platforms restricting third-party AI agent access is accelerating. It is documented across Amazon, JPMorgan, United Airlines, and others in Amazon, JPMorgan, and Delta Block AI Agents, Exposing Agentic Access Risk. This may draw regulatory attention to the asymmetry between first-party and third-party agent access.

Related Coverage

Corporate Policy2026-11-21

Signal Blocks OS-Level AI Agents, Exposing a Gap No App Can Fully Fix

Signal has restricted screenshots on Android and iOS 27 to prevent operating system AI agents, including Siri and Gemini, from reading users' on-screen messages. Signal president Meredith Whittaker described the gap as an existential threat to application-layer privacy and criticized Apple, Google, and Microsoft for not giving developers adequate tools to restrict AI data access. The fix is incomplete: Signal acknowledges it cannot block all access pathways, and the same protections are not yet available on macOS.

Corporate Policy2026-11-20

Meta's Muse Agent Leaks Other Users' Bank Balances and Signal Messages

Internal Meta employee reports, surfaced by Reuters, describe Meta's AI agent Muse surfacing other users' bank balances, personal messages, and other sensitive data to unintended recipients. In one reported case, Muse reminded a user to respond to a Signal message the user says she never consented to share. Meta attributed the incidents to employee testing of pre-release features not available to the public.

Corporate Policy2026-11-21

Apple's macOS Agent Plans Rewrite Endpoint Access Control Assumptions

Apple is reportedly developing a first-party AI agent for macOS capable of browsing the web, running commands, writing code, and operating across apps. Simultaneously, Apple plans to introduce friction in macOS 27.3 that discourages users from granting third-party agents full system access. Neither the first-party agent nor the macOS 27.3 controls are yet in force. Bloomberg reports the in-house agent is still in early development. A launch is possible no sooner than the end of 2027.