Apple's recent preview of macOS 27.3 beta 2 marks a significant shift in endpoint security for macOS devices. The update introduces new controls that allow Mac users to grant more granular access to data for apps, particularly those leveraging artificial intelligence (AI) agents. These changes are specifically designed to address the growing security risks posed by AI-driven applications, which often require extensive data access to function but can inadvertently or maliciously expose sensitive information.
The new controls focus on two key APIs: Device Control and Data Access, as well as Full Disk Access. These APIs are critical for managing how applications interact with system resources and user data. With the rise of AI agents—such as those used in productivity tools, virtual assistants, and automated workflows—there is an increasing need for tighter control over what data these agents can access and how they use it. This update is a proactive measure to prevent data exfiltration, unauthorized data collection, and potential misuse of AI capabilities by malicious actors.
For defenders, this is a call to action. Organizations must understand how these new controls work, how to configure them effectively, and how to monitor for signs of abuse. Failure to do so could leave sensitive corporate and personal data exposed to AI-driven threats.
Technical Analysis
Affected Products and Platforms
The changes are currently available in macOS 27.3 beta 2 and are expected to be included in the stable release of macOS 27.3. The affected components include:
- Device Control API: Manages access to hardware components such as cameras, microphones, and other peripherals.
- Data Access API: Controls access to user data, including files, contacts, calendars, and other sensitive information.
- Full Disk Access: A permission that allows apps to access all files on the system, often required by backup software, security tools, and now AI agents.
How the Vulnerability/Attack Works
AI agents, by design, often require broad access to system resources and user data to perform their functions. For example, an AI-powered virtual assistant might need access to emails, calendars, and files to schedule meetings or retrieve documents. However, this broad access can be exploited in several ways:
- Data Exfiltration: Malicious AI agents or compromised legitimate agents could exfiltrate sensitive data to external servers.
- Unauthorized Data Collection: AI agents might collect more data than necessary, violating privacy policies or regulatory requirements.
- Abuse of Full Disk Access: Apps with Full Disk Access can read, modify, or delete any file on the system, making them a high-value target for attackers.
The new controls in macOS 27.3 beta 2 aim to mitigate these risks by:
- Granular Permissions: Allowing users to grant access to specific data types (e.g., only contacts, not files) rather than broad categories.
- Transparency: Providing clearer prompts and logs about what data an app is accessing and why.
- Revocation: Making it easier for users to revoke access if an app is behaving suspiciously.
Exploitation Status
As of now, there are no confirmed reports of active exploitation of these specific controls. However, the risk is theoretical but credible, given the increasing use of AI agents in both legitimate and malicious contexts. Organizations should treat this as a proactive defense opportunity rather than a reactive measure.
Detection & Response
Sigma Rules
The following Sigma rules are designed to detect suspicious behavior related to AI agents and data access on macOS endpoints. These rules focus on process execution, file access, and network activity that may indicate abuse of the Device Control or Data Access APIs.
---
title: Suspicious AI Agent Process Execution
id: 550d9a23-0f72-4e3a-a8c5-1d2f3b4c5d6e
status: experimental
description: Detects execution of processes that may be AI agents requesting excessive data access.
references:
- https://attack.mitre.org/techniques/T1059/
author: Security Arsenal
date: 2026/11/15
tags:
- attack.execution
- attack.t1059.003
logsource:
category: process_creation
product: macos
detection:
selection:
Image|endswith:
- '/usr/bin/python3'
- '/usr/bin/osascript'
- '/usr/bin/automator'
CommandLine|contains:
- 'AI'
- 'agent'
- 'data_access'
- 'full_disk_access'
falsepositives:
- Legitimate AI development or testing
level: medium
---
title: Unusual Full Disk Access Requests
id: 7a3f1c82-9e4b-4d67-bc12-3e5a8f901234
status: experimental
description: Detects processes requesting Full Disk Access, which may indicate an AI agent attempting to access sensitive data.
references:
- https://attack.mitre.org/techniques/T1005/
author: Security Arsenal
date: 2026/11/15
tags:
- attack.collection
- attack.t1005
logsource:
category: process_creation
product: macos
detection:
selection:
CommandLine|contains:
- 'Full Disk Access'
- 'com.apple.security.tcc'
- 'tccutil'
falsepositives:
- Legitimate backup or security tools
level: high
---
title: AI Agent Network Exfiltration
id: 9b4e2f18-7c3a-4d5e-8f6a-2b1c3d4e5f6a
status: experimental
description: Detects network connections from AI agent processes to external IPs, which may indicate data exfiltration.
references:
- https://attack.mitre.org/techniques/T1041/
author: Security Arsenal
date: 2026/11/15
tags:
- attack.exfiltration
- attack.t1041
logsource:
category: network_connection
product: macos
detection:
selection:
Image|endswith:
- '/usr/bin/python3'
- '/usr/bin/osascript'
DestinationIp|cidr:
- '10.0.0.0/8'
- '172.16.0.0/12'
- '192.168.0.0/16'
filter:
DestinationIp|cidr:
- '10.0.0.0/8'
- '172.16.0.0/12'
- '192.168.0.0/16'
condition: selection and not filter
falsepositives:
- Legitimate cloud-based AI services
level: high
KQL (Microsoft Sentinel / Defender)
The following KQL queries can be used in Microsoft Sentinel or Defender to hunt for suspicious activity related to AI agents and data access on macOS endpoints. These queries leverage the DeviceProcessEvents and DeviceNetworkEvents tables.
// Hunt for AI agent processes requesting excessive data access
DeviceProcessEvents
| where FileName in~ ("python3", "osascript", "automator")
| where ProcessCommandLine has_any ("AI", "agent", "data_access", "full_disk_access")
| project TimeGenerated, DeviceName, FileName, ProcessCommandLine, InitiatingProcessFileName, InitiatingProcessCommandLine
| order by TimeGenerated desc
// Hunt for processes requesting Full Disk Access
DeviceProcessEvents
| where ProcessCommandLine has_any ("Full Disk Access", "com.apple.security.tcc", "tccutil")
| project TimeGenerated, DeviceName, FileName, ProcessCommandLine, InitiatingProcessFileName, InitiatingProcessCommandLine
| order by TimeGenerated desc
// Hunt for network connections from AI agent processes to external IPs
DeviceNetworkEvents
| where InitiatingProcessFileName in~ ("python3", "osascript")
| where RemoteIP !startswith "10." and RemoteIP !startswith "172.16." and RemoteIP !startswith "192.168."
| project TimeGenerated, DeviceName, InitiatingProcessFileName, InitiatingProcessCommandLine, RemoteIP, RemotePort, RemoteUrl
| order by TimeGenerated desc
Velociraptor VQL
The following Velociraptor VQL query can be used to hunt for suspicious processes and network connections related to AI agents on macOS endpoints.
-- Hunt for suspicious processes matching AI agent indicators
SELECT Pid, Name, CommandLine, Exe, Username, CreateTime
FROM pslist()
WHERE CommandLine =~ 'AI|agent|data_access|full_disk_access'
OR Exe =~ '/usr/bin/python3|/usr/bin/osascript|/usr/bin/automator'
-- Hunt for network connections from AI agent processes
SELECT Pid, Name, CommandLine, Exe, Username, CreateTime, RemoteIP, RemotePort
FROM netstat()
WHERE Name =~ 'python3|osascript'
AND RemoteIP !~ '10\.|172\.16\.|192\.168\.'
Remediation Script (Bash)
The following Bash script can be used to audit and harden macOS endpoints by checking for apps with Full Disk Access and revoking unnecessary permissions. This script is designed to be run on macOS 27.3 beta 2 or later.
#!/bin/bash
# Audit Full Disk Access permissions
LOG_FILE="/var/log/full_disk_access_audit.log"
echo "=== Full Disk Access Audit - $(date) ===" > "$LOG_FILE"
# List all apps with Full Disk Access
echo "Apps with Full Disk Access:" >> "$LOG_FILE"
sqlite3 /Library/Application\ Support/com.apple.TCC/TCC.db \
"SELECT client, auth_value FROM access WHERE service='kTCCServiceSystemPolicyAllFiles';" >> "$LOG_FILE"
# Check for AI agent processes
AI_PROCESSES=$(ps aux | grep -E 'python3|osascript|automator' | grep -v grep)
if [ -n "$AI_PROCESSES" ]; then
echo "Suspicious AI agent processes detected:" >> "$LOG_FILE"
echo "$AI_PROCESSES" >> "$LOG_FILE"
fi
# Revoke Full Disk Access for specific apps (example: revoke for com.example.maliciousapp)
# Uncomment and modify as needed
# sqlite3 /Library/Application\ Support/com.apple.TCC/TCC.db \
# "DELETE FROM access WHERE service='kTCCServiceSystemPolicyAllFiles' AND client='com.example.maliciousapp';"
echo "Audit complete. Review $LOG_FILE for details."
Remediation
To mitigate the risks associated with AI agents and data access on macOS endpoints, organizations should take the following steps:
- Update to macOS 27.3 or Later: Ensure all macOS devices are updated to the latest stable release (macOS 27.3 or later) to benefit from the new granular controls.
- Review and Restrict Full Disk Access: Use the new controls to review which apps have Full Disk Access and revoke permissions for any apps that do not require it. This can be done via:
- System Settings > Privacy & Security > Full Disk Access
- MDM Solutions: Use Mobile Device Management (MDM) tools like Jamf or Microsoft Intune to enforce policies.
- Implement Least Privilege for AI Agents: Configure AI agents to request only the minimum necessary permissions. For example, an AI assistant that only needs calendar access should not request Full Disk Access.
- Monitor for Suspicious Activity: Deploy the Sigma, KQL, and VQL rules provided above to detect and respond to suspicious behavior related to AI agents and data access.
- Educate Users: Train users to recognize and report unusual permission requests from AI agents or other applications.
- Leverage MDM for Enforcement: Use MDM solutions to enforce granular permissions and prevent users from granting excessive access to apps.
Official Vendor Advisory
For the latest updates and detailed guidance, refer to Apple's official security advisories and documentation:
CISA Deadlines
As of now, there are no CISA deadlines specifically related to this update. However, organizations should treat this as a high-priority proactive measure and implement the new controls as soon as possible.
Related Resources
Security Arsenal Penetration Testing Services AlertMonitor Platform Book a SOC Assessment vulnerability-management Intel Hub
Is your security operations ready?
Get a free SOC assessment or see how AlertMonitor cuts through alert noise with automated triage.