AI Sec Watch

AI-powered surveillance systems combining facial recognition (technology that identifies people by analyzing their faces), real-time tracking, and mass databases are being deployed globally to monitor and immediately enforce rules, with China operating over 600 million AI cameras and the US Department of Homeland Security rapidly expanding its use for monitoring immigrants, protesters, and journalists. These systems create 'chilling effects' (where people self-censor and conform out of fear of being watched), raising concerns about discrimination, lack of transparency, and threats to democracy, as the automation of surveillance and enforcement removes human judgment and accountability.

Researchers have discovered 'HalluSquatting,' an attack that exploits AI hallucinations (when AI systems generate false information or invent things that don't exist) to deliver malware at scale. Attackers pre-register fake repository and package names that AI coding assistants commonly hallucinate, then plant malicious code in those fake repositories so that when users ask their AI tools to clone or install resources, the AI may pull down the attacker's code and execute it.

Deutsche Telekom, a major telecommunications company, is transforming itself into an AI-native organization by embedding AI (artificial intelligence) throughout its operations rather than simply adding it to existing processes. The company is using generative AI (AI that creates text, code, or other content) in customer service, network operations, and voice communications, with over 50,000 employees actively using ChatGPT Enterprise and seeing a 546% increase in AI tool usage since early 2026.

According to Check Point's CTO, AI is transforming cybersecurity in two opposing ways: security companies can now scale their threat detection and defense operations dramatically (for example, using AI agents to make red teams, which test product security, about 20 times more efficient), but attackers are also using AI to launch phishing and malware campaigns faster and with less expertise. The main challenge organizations face is that AI agents behave unpredictably and require many system connections to function, which creates a larger attack surface (the total points where a system can be attacked) that is harder to protect than traditional deterministic systems (systems that produce the same output for the same input).

OpenAI attempted to explain how ChatGPT Work handles data across different platforms: cloud-based work syncs between web and mobile, while desktop work can access local files with permission, but conversations don't automatically appear across platforms. The company acknowledged this explanation was unsuccessful, suggesting the feature's data handling and synchronization behavior was confusing to users.

AWS has added OAuth support to its MCP Server (Model Context Protocol, a standard for connecting AI agents to external tools), allowing AI agents like Claude to access AWS services using the same login methods you'd use for the AWS console or command line. The service includes new security features like token revocation, dynamic client registration, and new audit logging in AWS CloudTrail (AWS's service that records all actions taken in your AWS account).

Meta released a new AI image generator called Muse that can create pictures of people by using photos from public Instagram profiles without notifying those users. Privacy advocates are warning Instagram users to check their privacy settings because the tool automatically uses public profiles for this image generation by default.

GhostApproval is a vulnerability affecting six major AI coding assistants (Amazon Q Developer, Claude Code, Augment, Cursor, Google Antigravity, and Windsurf/Devin Desktop) that allows attackers to escape sandboxes (isolated, restricted environments) by tricking the AI into accessing files outside the workspace while misleading the human reviewing the action. The attack exploits symbolic links (special files that act as shortcuts to other files or directories) combined with UI misrepresentation, where the confirmation prompt shown to the user hides dangerous information so they unknowingly approve harmful file access.

OpenAI is shutting down ChatGPT Atlas, a browser tool that could perform tasks automatically on a user's behalf, less than a year after launching it in October, with a shutdown date targeted for August 9th. The discontinuation is part of OpenAI's strategy to focus development efforts on productivity features and consolidate its products into a new desktop application called ChatGPT Work.