Weekly Cyber Threat Digest: Breaches, AI Exploits, and Critical Patches (April 27)

Top Attacks and Breaches

Vercel Breach via Connected App

Frontend cloud platform Vercel disclosed a security incident originating from a compromise at Context.ai. Stolen OAuth tokens allowed unauthorized access to Vercel through a linked application. The breach exposed employee information, internal logs, and a subset of environment variables. However, the company assured that the most sensitive secrets remained untouched.

Weekly Cyber Threat Digest: Breaches, AI Exploits, and Critical Patches (April 27) — Source: research.checkpoint.com

France Titres Identity Service Breach

France Titres, the authority managing identity and registration documents, detected a data breach on April 15. The incident potentially exposed names, birth dates, email addresses, login IDs, and in some cases physical addresses and phone numbers. A hacker later offered the purported agency data for sale on the dark web.

UK Biobank Health Data Exposed

UK Biobank, a major research organization, confirmed a breach after de-identified health data of 500,000 volunteers was advertised for sale on Chinese marketplaces. Officials stated that the listings were removed and believed unsold. In response, access was suspended, the research platform shut down, and download limits were imposed.

Bitwarden Supply-Chain Attack

Password manager Bitwarden suffered a supply-chain attack when a malware-tainted CLI release was published to npm on April 22. The company reported that 334 developers installed version 2026.4.0 during a brief window, potentially exposing credentials after a hijacked GitHub account was abused. Crucially, vault data remained unaffected.

AI-Powered Threats

Unreleased Anthropic AI Model Accessed

Researchers flagged unauthorized access to Anthropic's Claude Mythos Preview, an unreleased AI cyber model, through a third-party vendor environment. A small Discord group reportedly leveraged shared contractor accounts, API keys, and predictable URLs to reach the system. Anthropic said it is investigating and has not observed impact on core systems.

Bissa Scanner: AI-Assisted Exploitation Platform

Researchers observed Bissa Scanner, an AI-assisted exploitation platform using Claude Code and OpenClaw. It supported mass scanning, exploitation, and credential harvesting. The operation focused on exploiting React2Shell (CVE-2025-55182), scanning millions of targets, confirming over 900 compromises, and collecting tens of thousands of exposed environment files.

Google Antigravity IDE Sandbox Escape

Researchers highlighted a prompt-injection exploit chain in Google's Antigravity agentic IDE that enabled sandbox escape and remote code execution. The flaw abused a file search tool that ran before security checks, allowing attackers to convert a benign prompt into system compromise, even in Secure Mode. Google patched the vulnerability.

Vulnerabilities and Patches

Microsoft Out-of-Band Fix for Critical ASP.NET Flaw

Microsoft issued out-of-band fixes for CVE-2026-40372, a critical ASP.NET Core privilege escalation flaw rated 9.1. A bug in Data Protection versions 10.0.0 to 10.0.6 could let attackers forge cookies and antiforgery tokens, impersonate users, and gain SYSTEM-level access on Linux or macOS deployments.

Apple iOS Notification Services Bug Patched

Apple released fixes for CVE-2026-28950 in iOS and iPadOS, a Notification Services bug that could allow unauthorized access. Details were not fully disclosed, but users are urged to update their devices promptly.