Adil Islam

Daily Cyber Briefing — November 12, 2026

Attackers want the keys your pipelines and AI tools hold, and they are handing the legwork to agents of their own. Window roughly Nov 11, 7:00 a.m. to Nov 12, 7:00 a.m. CT. Sources are linked. Verification is marked on each item.

wfx-loot: hijacked GitHub accounts push a CI secret stealer into tens of thousands of repositories

Verified

Starting November 10, an attacker holding stolen GitHub credentials for 4,837 developers began opening pull requests titled "docs: dependency audit notes" in about 61,764 repositories, according to Aikido. The pull request is only a probe to confirm write access. Where it works, the attacker pushes a roughly 1,500-line stealer into an existing workflow, an npm postinstall hook, or a throwaway workflow branch, then reverts the change within a minute and deletes the workflow run and its logs. More than 10,000 of the probes landed in repositories the account doesn't own, reached through collaborator or organization access, and about 92% are still open.

The stealer takes GitHub Actions secrets, the workflow token, OIDC and runtime tokens, runner memory, cloud credential files, and tokens for AI providers and AI coding tools including OpenAI, Anthropic, Gemini, Hugging Face, Cursor, Copilot and Codex. It also tries to use OIDC against AWS, Azure, GCP, Vault and Kubernetes and plants cloud persistence. Aikido sees signs the payload ran in at least 1,806 repositories; SafeDep counted 8,093 repositories with malicious commits by early November 12, and StepSecurity says new probes were still appearing.

Why it matters: Developer tokens stolen in earlier supply-chain attacks are now being cashed in at scale, and a missing run number may be the only trace. Search your organization for wfx-docs branches and the listed commit messages, rotate every secret a touched repository could read, check AWS and Azure for new persistence, and restrict who can modify workflow files.

Sources: Aikido · SafeDep · StepSecurity · Help Net Security

Silent Lynx ran Central Asian espionage through a self-hosted Hermes AI agent

Verified

CloudSEK found an exposed server belonging to Silent Lynx, a group it links with medium confidence to Kazakhstan, holding stolen documents, tools and records of how the operator worked. Much of the work ran through Hermes, an open-source autonomous agent framework, configured as an offensive security expert, running mostly on DeepSeek models, with persistent memory, a task board for sub-agents, scheduled jobs and an imported library of offensive skills. The operator gave it instructions in Russian through a Telegram bot, and destructive commands were pre-approved.

The agent handled reconnaissance, data collection and cleaning up logs, and wrote the program that bulk-downloaded files. Using one railway employee's account, the group got into Tajikistan's national electronic document system and took more than 2,000 documents tied to the Defense Ministry, Reuters reported. CloudSEK also found a webshell on Uzbekistan's Smart Forestry platform and lures built in October for a campaign against Turkmenistan. The evidence does not establish a link to the Kazakh government.

Why it matters: Agent-run intrusions are no longer just a criminal tactic; regional espionage crews are using the same free frameworks defenders and hobbyists run. The tell is speed and volume from a single stolen account, so watch for one identity pulling far more documents than any person would.

Sources: CloudSEK · The Record · Reuters via Mint

Notion confirms Fireworks breach exposed Notion AI chat transcripts

Verified

Notion says the Fireworks AI breach reached Notion AI chat transcripts used for its Insights feature, affecting fewer than 1% of workspaces. Fireworks first told Notion on November 2 that only synthetic testing data was accessed, then on November 6 that customer data may have been, and confirmed the dataset on November 8. Most of the transcripts carry no label tying them to a user or workspace. Notion disconnected from Fireworks, disabled Insights, stopped serving open-weight models through it, and tells affected customers to rotate any passwords or keys they pasted into Notion AI.

Why it matters: Your data can sit with an AI inference vendor you never signed a contract with, and "zero data retention" defaults don't cover features that store content. Ask every SaaS vendor which model providers process your prompts and which of those features keep them.

Sources: Notion advisory · SecurityWeek · VentureBeat

New N-central exploits beat N-able's two-day-old hotfix

Verified

The researcher behind the Able2Shell disclosures published two more exploit sets that work against N-central 2026.4 Hotfix 2, the fix N-able shipped November 10. 4EverAble2Shell is an unauthenticated chain to root on the server, which N-able says needs the non-default Legacy API turned on and exposed; hosted N-central is not affected. Agent2Shell is more than ten issues in how the server talks to Windows Agents and Probes, and two chains run commands as SYSTEM on managed endpoints if the attacker knows the appliance ID. There is no fix yet, and watchTowr and Previdian saw probing within hours.

Why it matters: An RMM server reaches every machine it manages, so one exposed setting can turn an MSP into a launch pad. Disable the Legacy API now and keep on-premises N-central off the internet until N-able ships a fix for the agent chains.

Sources: Huntress · The Hacker News · Cybersixt

SCKit worm plants rogue ScreenConnect clients across 48 organizations

Verified

Starting around 17:47 UTC on November 11, an operator used the Run Command feature on cloud-hosted ScreenConnect panels to install attacker-controlled ScreenConnect clients and a 16-file toolkit called SCKit on at least 288 endpoints across 48 organizations, Huntress says, all within 90 minutes. SCKit waits for a technician to connect and is built to capture that session and spread to everything the technician's panel can reach. Huntress traced initial access to the third-party RMM+ extension, whose versions 1.0.31 and 1.0.32 accepted a blank access key; ConnectWise disabled five extensions across about 400 partner environments.

Why it matters: Don't triage a suspect machine over ScreenConnect, because connecting is how this spreads. Use EDR live response or another out-of-band tool, remove third-party extensions you don't need, and revoke all sessions from a clean device.

Sources: Huntress · ConnectWise · BleepingComputer

Microsoft's AI kernel scan: 59 bug reports, 33 CVEs, root exploits for eight

Verified

Microsoft's FORGE Lab ran MDASH, an AI-agent scanning pipeline, across Linux kernel subsystems. It produced 26,502 raw findings in two weeks, which months of human work cut to 59 upstream reports, 33 CVEs and 38 merged fixes; Microsoft says 24 of the CVEs are rated High or above and researchers built root exploits for eight. Microsoft stresses that people still did root-cause analysis, patches and maintainer coordination.

Why it matters: AI now finds bugs faster than anyone can triage them, so validation and patching capacity is the real limit. Expect a steady rise in kernel and open-source CVEs and budget patch cycles accordingly.

Sources: Microsoft Security Blog · CSO Online

Also on the wire

  • Maersk confirmed data was taken from a legacy system it inherited in an acquisition, as the WhiteArmor group demands $5 million; operations are not disrupted. Maersk
  • Microsoft Threat Intelligence says attackers are planting GSocket on Linux servers to keep access after the original web application flaw is fixed. Microsoft Threat Intelligence on X
  • XBOW says it helped find 14 of the 36 flaws in SAP's November patches, including a CVSS 9.9 code injection in NetWeaver Knowledge Management. XBOW on X
  • Google Cloud warned of three AMD EPYC Turin flaws that can leak data inside VMs on several instance families. The Hacker News on X
  • Node.js fixed 16 security flaws, including an HTTPS bug that can expose TLS private keys under specific conditions. The Hacker News on X
  • Elliptic found North Korea laundering the $6.4 billion Binance theft through meme coins it mints for about $90 each. International Cyber Digest 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