IT Daily Roundup – October 6, 2026
1. Microsoft Exchange: CVE-2026-96940 (CVSS 8.8) – Authenticated Attackers Can Read Colleagues' Mailboxes
Microsoft patched a high-severity vulnerability CVE-2026-96940 (CVSS 8.8) in Exchange Server that allows an authenticated attacker to gain access to other users' mailboxes within the same organization. The flaw lies in insufficient authorization: an attacker with valid credentials can escalate privileges and silently read emails, attachments, and internal communications belonging to any user. Affected versions include Exchange Server Subscription Edition RTM, Exchange Server 2016 Cumulative Update 23, and Exchange Server 2019 Cumulative Updates 14 and 15. Microsoft has not yet observed active exploitation, but the "Exploitation More Likely" assessment indicates it is only a matter of time. Exchange Online has been patched automatically and requires no action from administrators.
Why it matters: While most SMEs today rely on Microsoft 365 (Exchange Online), a significant number of organizations still run Exchange on-premises—for compliance reasons, legacy software integration, or historical choice. A vulnerability that compromises one mailbox in a large corporation can, in a smaller firm, mean access to the director's, accountant's, or lawyer's correspondence—with no trace left behind. If your organization runs Exchange on-premises, contact your IT partner and verify the patch has been applied today. If you're unsure whether this affects you, that's a good question for your IT administrator.
2. 543,000 Active Credentials Exposed: Research Reveals Widespread GitHub Leaks, Some Dating to 2009
Truffle Security analyzed The Stack v3—a public code snapshot assembled for training large language models—and found 543,699 still-active credentials across 224 million GitHub repositories. The exposed material included 69,041 Google Cloud service account keys, 51,067 MongoDB connection strings, 33,343 Google API keys, 31,374 Gemini keys, and tens of thousands of database credentials for PostgreSQL and MySQL, as well as npm and GitHub tokens. The duration of exposure is particularly alarming: the median was 784 days—nearly two and a half years. One database credential remained valid against an Erlang server where it was last embedded in June 2009—over 16 years ago. Of 12,985 PostgreSQL connection strings tested, 11,465 (88%) were still functional.
Why it matters: This isn't just a problem for large software companies with thousands of developers—it affects any organization where someone has ever managed code or scripts in a GitHub repository. A common scenario: a developer tests a connection to a production database, commits a file containing the connection string, the repository is "internal only"—then a year later it's accidentally made public or indexed by an AI training tool. Recommended steps: (1) install GitHub Advanced Security or the free git-secrets tool to detect credentials before they're committed; (2) set up regular rotation schedules for API keys and database passwords; (3) for existing repositories, run an audit using tools like TruffleHog or GitLeaks. Any committed credential must automatically be treated as compromised—removing it from git history is not enough, since GitHub indexes and AI training datasets may retain it.
3. Apple Tightens macOS Full Disk Access: AI Agents Must Receive More Explicit Consent
Apple announced planned changes to the Full Disk Access permission on macOS, which currently allows apps to read virtually all files on a Mac including emails, messages, and browsing history. The reason is growing concern about autonomous AI agents—Apple specifically named Meta Muse and OpenAI Dots. The company stated: "As AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially." New controls will require more explicit user consent before granting Full Disk Access; Apple has not announced a specific timeline for the rollout. The changes will affect not only AI applications, but also legitimate backup and system utilities (such as Carbon Copy Cloner) that currently rely on Full Disk Access.
← All postsWhy it matters: This news comes shortly after Google was found testing full Gemini access to Mac files—and Apple is responding in the opposite direction: raising barriers rather than lowering them. For organizations where employees work on Macs, this is a dual signal: (1) IT teams should review which applications currently have Full Disk Access granted and whether it is genuinely necessary; (2) any AI tool requesting full disk access represents a security risk unless the vendor transparently explains what it does with the data. Backup tools and system utilities that legitimately require Full Disk Access should be explicitly approved by IT administration—not left to the discretion of an employee who clicks "Allow" without thinking.