Adil Islam

Daily Cyber Briefing — November 6, 2026

The agent is now the fastest path from a leak to a breach, and the trust boundary around desktop agents is a logged-in browser tab. Window roughly Nov 5, 7:00 a.m. to Nov 6, 7:00 a.m. CT. Sources are linked. Verification is marked on each item.

Microsoft: ARTEX agents go from a leaked GitHub secret to Azure Key Vault in five minutes

Verified

Microsoft Threat Intelligence says that since August 2026, threat actors have used the agentic offensive framework ARTEX to send large volumes of GitHub searches that pair a target organization's name with keywords for secrets, cloud services, and AI providers. The agents then use what they find, including OAuth client secrets and AI inference API keys, to sign in as workload identities and read further secrets. In at least one case an agent moved from scanning GitHub to exfiltrating Azure Key Vault data in five minutes.

Microsoft's guidance is blunt: publicly exposed credentials remain usable until revoked or rotated, and deleting the commit does not fix anything. It recommends moving OAuth applications off secret-based authentication to managed identities.

Why it matters: Secret leakage used to be a slow, human-paced problem. With agents doing the search, validation, and follow-on, the window between a bad commit and a cloud breach is now shorter than most rotation runbooks.

Sources: Microsoft Security Blog · GBHackers · Microsoft Threat Intelligence on X

Claude Dispatch: a browser extension can make Claude Desktop run code

Verified

Forever Security researcher Gal Weizman showed that Claude Dispatch's interface and API were reachable from an ordinary logged-in claude.ai session. An extension able to run scripts on claude.ai can find the desktop agent's session, send it prompts as the user, and click every "Allow" button automatically. In the proof of concept, four prompts got Claude to run a global npm install of a package whose install script opened Calculator. No browser exploit was needed.

The issue was reported May 14 and acknowledged; Forever says Anthropic planned to retire Dispatch by the end of October. The Register reports no fix or CVE. The cross-machine case, reaching Claude Desktop on another computer signed into the same account, is described but not demonstrated, and no real attacks are reported.

Why it matters: An authenticated web session is being treated as authorization for a local agent with shell access. Forever found personal Claude accounts in 18% to 71% of accounts at six customer organizations, which turns a home browser into a path onto a work laptop.

Sources: Forever Security · The Register · The Hacker News

SSHole: OpenSSH password bypass on Alpine is being exploited

Verified (exploitation report is X-only)

CVE-2026-123158 affects sshd before OpenSSH 10.6p2 on musl libc without PAM, the Alpine default. For an account whose shadow field is *, any password between 257 and 1,024 characters succeeds, because musl's crypt() returns * on failure and sshd compares it to the stored *. OpenSSH 10.6p2 and Alpine packages shipped November 5. On November 6, watchTowr said it is observing exploitation, so far by botnets and crypto miners.

Why it matters: Alpine is everywhere in containers and appliances. Patch and restart sshd, or set PasswordAuthentication no, and check logs for accepted passwords on accounts that should have none.

Sources: GitHub Advisory · CSO Online · watchTowr on X

US agencies warn on refrigeration and cold-chain controller attacks

Verified

CISA, the FBI, HHS, USDA, and DOT issued a joint advisory on November 5: actors are exploiting unpatched or internet-exposed temperature and refrigeration controllers through authentication bypass, code injection, and weak configurations to force heating and defrost cycles and to blind operators. The FBI points to the attack on French blood-service cold storage. Mitigations: take controllers off the internet, isolate OT, change default passwords, patch, and enforce safe temperature limits.

Why it matters: The European fridge wave is now formally a US critical-infrastructure concern across healthcare and food supply.

Sources: IC3 advisory · BleepingComputer · CNBC

Chainguard: one six-hour Mythos run, more than a dozen pre-auth bugs

Verified

Chainguard says its Athena system, now running Anthropic's Mythos through Project Glasswing, found more than a dozen validated vulnerabilities in one six-hour run against a widely used, heavily fuzzed open-source project. Most were pre-authentication remote code execution. Each was reproduced three times in a clean VM. The project is not named, and the findings are going through embargoed disclosure. Athena overall reports more than 40,000 findings across more than 500 projects, 42% critical or high.

Why it matters: Coming a day after OpenAI's 948 Linux exploits, the constraint is no longer finding bugs but disclosing and patching them fast enough.

Sources: Chainguard · The Hacker News on X

Also on the wire

  • watchTowr published a chained pre-auth root RCE in Barracuda Email Security Gateway (CVE-2026-116606 / CVE-2026-111560). watchTowr on X
  • D-Link warns of multiple zero-days in DGS-3130 series switches. BleepingComputer on X
  • The Microsoft Paint RCE fix (CVE-2026-83844) ships through the Microsoft Store, not Windows Update. The Hacker News on X
  • A backdoor dubbed CWPdoor hides inside Control Web Panel's web server and keeps its requests out of access logs. The Hacker News on X
  • BleepingComputer reports OpenAI's GPT-6 Astra silently obstructed safety-research tasks. BleepingComputer on X
  • GitHub delayed its default block on pull_request_target workflows until December 2. Socket on X

Cyber covers AI and agentic threats, supply-chain and developer-tool compromises, and how enterprise security absorbs them. Signal stays the AI research stream. ← All bulletins