The Tuesday Briefing — Nov 10, 2026

The Big Picture
One security company counted 1,379 live passwords and keys sitting in public code folders, left behind by a single "don't ask again" button in an AI coding tool. Meanwhile, the company behind a tool many small IT providers use to run client computers took its hosted service offline over 33 publicly posted security holes. This week's theme: the tools your team and your IT provider use to get work done are the doorway.
This Week's Top 5
1. A "Don't Ask Again" Click Left Passwords in Public Code Folders
What happened: Security firm Apiiro found that 1,122 of 104,185 public copies of a Claude Code settings file contained apparent live secrets. The file records commands a developer approved with "Yes, and don't ask again," so passwords and keys typed into those commands get saved in plain text. Later analysis counted 1,379 credentials in one scan and up to 2,539 in a broader one. Anthropic acknowledged the issue, and a full fix was still pending as of Nov 8.
Why it matters to your business: Deleting the file later does not remove it from the project's history or from copies other people made. Anyone who finds it can use the keys.
What to do: Ask your developers to search your code folders for files named .claude/settings.local.json, change any password or key found in them, and add that file to the folder's ignore list. Read the report.
2. The Tool Many IT Providers Use to Manage Client Computers Went Offline
What happened: N-able took its hosted N-central service offline and told customers running it on their own servers to shut those servers down. A researcher had publicly posted 30 or more security holes (reported as 33), with working attack code, and no fixes or workarounds had been announced at the time.
Why it matters to your business: N-central is software used by IT service companies to manage many clients' computers at once. If your IT provider uses it, a break-in there could reach your machines too.
What to do: Ask your IT provider today whether they use N-central, what they have done in response, and whether your backups can be reached by their management tool. More on the shutdown.
3. Attackers Are Using AI Agents to Break Into Companies Fast
What happened: CrowdStrike says a China-linked group it calls Vapor Panda used an AI hacking platform against more than 200 organizations in at least 14 countries over 11 days. Separately, Acronis described a 12-agent attack system that reached victims in 11 countries by exploiting basic mistakes like hardcoded keys and default admin passwords. Microsoft linked a related AI tool, ARTEX, to attacks on banks. This week's new detail: an exposed GitHub secret was reportedly turned into a stolen Azure Key Vault in about five minutes.
Why it matters to your business: The weaknesses being used are ordinary ones: default passwords, exposed admin pages and leaked keys. A leaked key can be abused within minutes.
What to do: Ask your IT provider to scan for exposed admin logins and default passwords, and to confirm they can replace a leaked key the same day. Acronis report.
4. A "Verified" Software Release Still Carried a Key Stealer
What happened: Attackers used Ethersphere's own release process to publish a malicious version of its bee-js software library (13.3.0) and Swarm Desktop 0.56.2. The bee-js release carried a valid "provenance" stamp, which says where software was built, yet it stole node keys and planted backdoors.
Why it matters to your business: A trusted-looking stamp does not prove the software is safe. If your developers add new versions automatically, bad updates can arrive the same way.
What to do: Ask your developers to review new software versions before adopting them, rather than pulling the newest release automatically. Read more.
5. Update Chrome, Android and Your Remote-Access Gear Now
What happened: Google patched the eighth Chrome flaw of 2026 that attackers were already using (Chrome 157.0.8103.50/.51). Android's November update fixed a zero-click flaw (CVE-2026-58868), meaning no tap is needed. SonicWall's VPN flaw (CVE-2026-19868) lets attackers get around the second sign-in step, and the US cyber agency CISA lists it as exploited. Barracuda email security gateways also have two flaws that attackers are using to gain full control.
Why it matters to your business: These are products many small firms run or carry in their pockets, and attackers are already using these flaws.
What to do: Check that Chrome is on version 157.0.8103.50 or newer and that Android phones show a patch level of 2026-11-05 or later. If you use a SonicWall VPN or Barracuda email gateway, ask your provider to confirm both are updated. Chrome details.
Quick Hits
-
OpenSSH, a common remote-login tool, had a flaw nicknamed "sshole." It was reportedly attacked within 24 hours of disclosure. Fixed version: 10.6p2. Reports disagree on how severe it is, so ask your provider to update.
-
France blamed an Iranian military cyber unit for sabotaging refrigeration controllers at about 700 retail sites, with some incidents reportedly causing fires. Any equipment reachable from the internet should be hidden behind a secure connection.
-
Anthropic told enterprise customers that a third-party security firm accessed account metadata (names, emails, org IDs, plan tiers, usage and spend). Anthropic says access was disabled and deletion confirmed. Expect more convincing phishing emails and watch for them.
-
Google warned that "LLMjacking" attackers steal cloud keys to run up AI bills, some above $100,000 per day. Set spending caps on every AI account.
-
Researchers found 575 or more malicious add-on "skills" for the OpenClaw AI agent that delivered password stealers and other malware. Only install add-ons you have approved in advance.
-
The UK is discounting Cyber Essentials certification by 75% for eligible small businesses.
-
Vanderbilt University was hit by ransomware that reportedly started on a personal device. Staff devices need to meet the same security rules as company ones.
-
Fireworks AI disclosed a breach after attackers compromised a third-party cloud provider and reached development machines. Ask your vendors how they limit what outside providers can reach.
One Thing to Do This Week
Ask every developer, contractor or IT person who uses an AI coding tool to search for leftover passwords in their code folders, then change anything they find. This week's story shows that one "don't ask again" click can write a live password into a file that later gets shared publicly. Here is how: have them search each project for a file called .claude/settings.local.json and open it. Look for anything that resembles a password, key or token. Change each one in the service that issued it first, because deleting the file does not cancel a copied key. Then add that file name to the project's ignore list so it is not shared again. If your team uses other AI coding tools, ask them to check those settings files too. Finally, ask whether anyone has turned on a secret-scanning feature, which can catch this automatically.
Worth Reading
-
BleepingComputer: Google LLMjacking attacks rack up $1 million in cloud AI charges. How stolen cloud keys get used to run up AI bills.
-
BleepingComputer: N-able takes cloud platform offline after 30-plus zero-days published. What happened to the IT management tool and what to ask your provider.
-
The Record: France blames Iran for fridge cyberattacks. The attribution behind the refrigeration-controller sabotage.
-
BleepingComputer: Hackers exploit "sshole" OpenSSH flaw to breach Linux systems. What the remote-login flaw allows and who is at risk.
Related Posts
The Tuesday Briefing — Nov 17, 2026
Weekly security intelligence for SMBs. Top threats, quick hits, and one action to take now.
The Tuesday Briefing — Nov 3, 2026
Weekly security intelligence for SMBs. Top threats, quick hits, and one action to take now.
The Tuesday Briefing — Oct 27, 2026
Weekly security intelligence for SMBs. Top threats, quick hits, and one action to take now.