Cyber News Bytes: GitHub Breach, Microsoft Emergency Patches, & Anthropic's Latest Milestone

Another crazy week in cyber with headlines from Anthropic, GitHub, Microsoft, and more!

Four stories shaped the cyber conversation this week, and the through line is the same one we keep landing on lately: the speed of attack is outrunning the speed of defense.

1. Anthropic's Project Glasswing has uncovered over 10,000 critical vulnerabilities in a single month

Anthropic published its first progress report on Project Glasswing this week.

The initiative uses an unreleased frontier model called Claude Mythos Preview to scan some of the most critical software in the world.

In just one month, around 50 partner organizations have collectively flagged more than 10,000 high or critical severity vulnerabilities using the model.

Of those, 1,726 have been validated as true positives, and only 97 have been patched so far.

Cloudflare alone reported finding 2,000 vulnerabilities across its critical systems.

Mozilla used Mythos to patch 271 vulnerabilities in Firefox 150, more than ten times the number found in the previous release.

Why it matters: The bottleneck in cybersecurity is officially shifting from finding flaws to fixing them, and the open source ecosystem is struggling to keep up with the new pace of discovery.

Read more at The Hacker News

2. GitHub confirmed a breach after attackers stole 3,800 internal repositories through a poisoned VS Code extension

A threat group called TeamPCP gained access to GitHub's internal systems by way of a malicious version of the popular Nx Console extension.

The compromised version was published to the VS Code Marketplace on May 18 and was live just long enough to harvest credentials from developer machines that updated during that window.

One of those machines belonged to a GitHub employee.

The attackers used that single foothold to exfiltrate roughly 3,800 internal repositories, then listed the data for sale on a cybercrime forum for over $50,000.

GitHub says no customer repositories, enterprise accounts, or user data were affected.

The same campaign has been linked to recent breaches at Grafana Labs, OpenAI, and Mistral AI.

Why it matters: Developer tooling has become a front door to enterprise breaches, and most security teams have almost no visibility into what extensions or packages are running on their developers' machines.

Read more at The Hacker News

3. Microsoft rushed out emergency patches for two actively exploited Defender zero-days

On May 21, Microsoft pushed out-of-band patches for two Windows Defender vulnerabilities already being used in live attacks.

The flaws are tracked as CVE-2026-41091 and CVE-2026-45498.

The first allows a low-privileged attacker to escalate to full SYSTEM-level control on a Windows machine.

The second can silently disable Defender's threat signature updates without alerting users or administrators.

Both vulnerabilities, publicly known as RedSun and UnDefend, were disclosed without coordination by a researcher who goes by Chaotic Eclipse.

CISA added them to its Known Exploited Vulnerabilities catalog on May 20 and gave federal agencies until June 3 to patch.

Why it matters: When the antivirus itself becomes the vulnerability, every endpoint relying on Defender becomes the target instead of the shield.

Read more at BleepingComputer

4. A new BitLocker bypass called YellowKey lets attackers crack encrypted Windows drives

Microsoft published mitigations this week for a publicly disclosed flaw now tracked as CVE-2026-45585.

The vulnerability, dubbed YellowKey, allows an attacker with brief physical access to a Windows machine to bypass BitLocker disk encryption entirely.

It works by abusing trusted behavior in the Windows Recovery Environment to drop into an unrestricted command shell with the drive already unlocked.

Affected systems include Windows 11 versions 24H2, 25H2, and 26H1, plus Windows Server 2025.

The same researcher behind RedSun and UnDefend, Chaotic Eclipse, published a working proof of concept before Microsoft had a fix ready.

Switching BitLocker from TPM-only to TPM plus PIN mode blocks the attack.

Why it matters: "BitLocker enabled" has been a comforting checkbox for years, but stolen laptops and unattended machines just became a much bigger problem than most organizations planned for.

Read more at The Hacker News

This was a week of trust running out.

The pattern keeps repeating because the tools we rely on most are also the ones we examine the least, and attackers have learned to walk in through the same doors our developers use every day.

Stay patched, stay skeptical! Talk soon.

Sandra | www.withcybersecurity.com