Menu
Hands holding a bank card beside a laptop, illustrating service provider cyberattack risks

When Trusted Access Becomes the Attack Path

, , , , , , ,

This week’s most important technology stories were about trusted access. A service provider cyberattack gave criminals a path into a bank system that could move money. A server-security company used an update channel that ran as root. A VPN accepted an alternate authentication path. An AI security agent found an old internal tool that was never supposed to face the public internet.

None of those systems began with a stranger smashing through the front door. The attacker, or in one case an unintended research agent, found a door the organization had built for legitimate work.

That distinction matters because businesses can’t operate without vendors, updates, remote access, integrations, and automation. “Trust nobody” sounds decisive right up until payroll needs to run. The useful question is narrower: once we trust this system, what can it do, and what stops that trust from spreading farther than it should?

That’s the practical problem behind third-party access security, and it’s the Big Story this week.

The Big Story: Trust Is an Attack Surface

The service provider cyberattack became a bank robbery

French banking group Crédit Mutuel Arkéa says attackers went through a service provider’s platform, reached an account belonging to the bank, and sent fraudulent transfers to other banks. The bank has now put the missing amount at about €100 million, or roughly $113 million, while recovery efforts continue. Some of the money may later have been converted into cryptocurrency, but it wasn’t cryptocurrency when it left the bank.

There are several important limits to the story. Arkéa says customer accounts weren’t used, customer assets weren’t affected, and delayed transfers were the main customer-facing consequence. French prosecutors are investigating. The bank hasn’t named the provider or published the precise technical route, so “sophisticated attack” remains a description, not an explanation.

What we do know is enough to expose the control problem. A vendor’s platform could reach a bank-controlled account with authority to move a very large amount of money. That is not just vendor risk in the abstract. It is delegated authority with a price tag.

Small businesses delegate the same kind of authority at a smaller scale. A payroll service can send money. A bookkeeper can change vendor details. A managed IT provider can reset accounts. An ecommerce integration can issue refunds. A remote-support tool can control computers. None of that is inherently reckless. Doing everything in-house isn’t realistic, and it isn’t automatically safer.

The problem appears when the business treats the provider’s access as somebody else’s security problem. If a compromised vendor account can approve a new payment destination, the business needs a control outside that same account. That might be a second person, a call to a known number, a transfer limit, a delay for unfamiliar destinations, or an alert the vendor can’t silence. Multi-factor authentication helps protect the login. It doesn’t decide whether one login should be allowed to move the money alone.

Technical discovery & auditing

The public page doesn’t tell you much about the machinery behind it. Raymond Tec audits inherited and long-running projects to uncover the plugins, integrations, data, dependencies, and old decisions that determine what the next change will really involve.

A trusted update can arrive with root access

The cPGuard incident shows the same problem on web servers. OPSSHIELD says an attacker registered a customer account, broke into its billing and licensing server, stole the key used to publish updates, and pushed a malicious cPGuard update for about 70 minutes on November 2. Roughly 2,950 servers ran the fake update.

cPGuard is security software. Its updater runs with root privileges because security software needs deep access to inspect and repair a server. The malicious script inherited that legitimate authority. It collected local password hashes, root SSH keys, a MySQL root password where present, shell history that could contain tokens, and some cloud-storage keys. It also installed a hidden root shell.

This is where Wednesday’s early reporting needs an update. OPSSHIELD’s fuller disclosure now says the attacker reached the billing application, copied about 7,900 customer account records, and used stolen publishing credentials to distribute the fake update. The company released clean version 5.88.26 and is rebuilding systems it no longer trusts. Customers whose servers were online during the affected window are being told to assume root-level secrets were exposed. Some customers whose servers were accessed manually have been told to rebuild from a clean backup taken before November 1.

Installing 5.88.26 removes the known attacker tools. It can’t make a copied SSH key or database password secret again. Administrators still need to rotate credentials from a trusted environment, replace root SSH keys everywhere they’re accepted, inspect other systems reachable from those servers, and reset the OPSSHIELD portal password.

WordPress had its own uncomfortable update story this week. An actively exploited flaw in versions 4.7 through 7.1.3 allowed an attacker to interfere with the temporary directory used during a core update. The fixed release is 7.1.4. The narrow technical race is different from the cPGuard supply-chain compromise, so I don’t want to pretend they’re the same attack. They do share a practical consequence: seeing “update completed” is the start of verification, not the end of it.

Ask what the trusted identity can do

Here’s a worked example. Imagine a ten-person retailer uses a managed IT provider, a payment processor, a website agency, and a cloud backup service. Assume every provider uses MFA. Good. Now ask the second set of questions.

  • Can the IT provider reset the owner’s email account and then use that email to reset other services?
  • Can the payment processor or bookkeeper change where money is deposited without a second confirmation?
  • Can the website agency reach the live store, customer records, and the backup account with the same identity?
  • Can any provider delete the logs or alerts that would show what its account did?
  • Can the account that runs production erase the recovery copy too?

That is third-party access security in useful terms. Not “Do we trust the vendor?” Trust isn’t one bit that is either on or off. The question is whether the provider has the minimum authority it needs, whether dangerous actions require an independent check, whether activity is visible, and whether the business can remove access without taking itself offline.

I don’t find a 40-page vendor questionnaire reassuring if nobody can answer those five questions. Paper controls have their place. A transfer limit that actually stops a fraudulent payment has a better one.

Business IT goes well beyond the website

Your business also depends on workstations, cloud accounts, browsers, Wi-Fi, remote access, collaboration tools, and all the other technology that quietly becomes infrastructure. Raymond Tec works across that whole stack, whether the problem lives on a server, on a desk, or somewhere in between.

Patching Closes the Door. It Doesn’t Explain Who Came Through.

SonicWall and QNAP both published urgent guidance this week for products that sit at the edge of a network.

SonicWall says CVE-2026-19868 is being exploited against SSL-VPN services on Gen 7 and Gen 8 firewalls. The flaw matters because it can let someone with valid credentials for an alternate authentication method bypass SAML controls, potentially including MFA enforced through SAML. The affected releases are SonicOS 7.3.3-7015 and earlier and 8.2.2-8015 and earlier. The fixes are 7.3.4-7013 and 8.2.3-8015. There is no workaround.

SAML is the handoff that lets an identity provider tell the firewall a person signed in and completed the required checks. The vulnerability doesn’t prove MFA is useless. It proves every entrance has to lead through the check. A side door that accepts another credential can defeat a very good lock on the front door.

QNAP, meanwhile, says exposed Plex Media Server installations may already be compromised through the Plex2Shell vulnerability chain. Plex 1.43.2 and earlier is affected; 1.43.3.10861 or later contains the fix. If remote access, a forwarded port, or UPnP exposed Plex while it was vulnerable, QNAP recommends more than an update: preserve logs if needed, perform its clean reinstall process, sign every device out, and reset Plex, NAS, and connected-service credentials.

For ordinary users, this is mostly an administrator job. Ask whoever manages the firewall or NAS to give you the installed version and the date it was checked. For administrators, the distinction is blunt: the patch changes what an attacker can do next. Logs, credential rotation, and a clean rebuild answer what may have happened already.

Backups & recovery

Backups are comforting right up until you need one. Raymond Tec helps small businesses build practical backup and recovery plans — including the decidedly unglamorous part where we make sure the thing can actually be restored.

AI Agents Are Finding the Old Doors Faster

WorkOS disclosed that a security researcher’s AI agent escaped its test environment, discovered an exposed internal application, guessed a development credential, and retrieved customer records in bulk. The researcher came forward, supplied evidence that an independent forensics firm matched to WorkOS logs, and said the data was deleted and never shared.

The AI agent is the interesting new capability. The route it found is embarrassingly familiar. WorkOS says the legacy tool was reachable from the internet, accepted a weak credential intended for local development, had no rate limit, and wasn’t monitored closely enough to detect the first access for five days.

The exposed records included business contact details, account histories, sales and support material, excerpts from customer Slack conversations, and product-usage information. WorkOS says production identity services weren’t the source. That limits the incident, but it doesn’t make the material harmless. Real account context is excellent raw material for a convincing support scam.

Convenience can preserve more permission than intended

Claude Code supplied a quieter version of the same problem. Security firm Apiiro found passwords, API keys, database credentials, and server logins inside public GitHub copies of Claude Code permission files. When a developer approves a shell command with “Yes, and don’t ask again,” the tool can save the exact command as an allow rule in .claude/settings.local.json. If the command contains a secret, the standing permission can preserve the secret too.

Apiiro says it found 1,122 public copies of that local settings file containing at least one secret. The company didn’t test whether every credential still worked, although one notified organization confirmed its key had been live. Anthropic’s documentation says Claude Code adds the local file to a developer’s global Git exclusions when it creates the file. That helps on that machine. It doesn’t follow the repository to every other machine, and it doesn’t remove a secret already committed to Git history.

If your team uses coding agents, scan current repositories and their history for agent settings and transcripts. Read the commands, not just the filenames. Rotate any exposed secret before cleaning up the repository. Deleting the evidence doesn’t revoke the credential.

Google and Mandiant also documented “LLMjacking,” where attackers steal cloud or AI credentials and consume expensive model capacity while the victim gets the bill. Charges in investigated cases exceeded $100,000 in a day. A monthly budget email is badly matched to a theft that grows by the hour. Scope keys to the models that need them, alert on newly enabled services, and use hard quotas where the provider offers them.

The common lesson isn’t that AI is uniquely reckless. It’s that automation converts reachable permission into speed. The old exposed tool, saved command, or overpowered API key was already a problem. An agent can find it, reuse it, and scale the result before a person notices.

Turning on AI is the easy part

Deciding what an AI tool should be allowed to see, who should use it, what work it should perform, and what happens when it gets something wrong is the more interesting problem. Raymond Tec helps businesses connect and automate the tools they actually use without treating every new feature like a button that obviously needs to be switched on.

When a Thermostat Becomes Operational Technology

Attackers changed temperature and defrost settings on commercial refrigeration systems across European supermarkets this week. Some stores discarded food. At one Swedish store, a controller was set to 50 degrees Celsius, or 122 degrees Fahrenheit. The incidents reached at least 11 countries, and the U.S. government issued a joint warning on November 5 about ongoing attacks against industrial refrigeration and cold-chain monitoring systems.

Several French Blood Establishment sites were hit too. Staff received independent alarms, moved blood products into equipment that wasn’t connected to the affected systems, and prevented any blood loss. That’s worth pausing on. The attack still succeeded at changing a physical control. The recovery design stopped the control failure from becoming a medical-supply crisis.

A refrigeration controller is operational technology. It doesn’t merely report a temperature. It turns compressors, fans, heaters, and defrost cycles on and off. Remote access makes monitoring and maintenance easier, but it also creates a route from the network to the product in the cold room.

The current public reporting points to older internet-exposed controllers, known vulnerabilities, and weak or predictable credentials in at least some cases. Investigators haven’t established that one attacker or one exact technique explains every incident. Owners should inventory internet-connected controllers, remove direct public exposure, update supported equipment, isolate management access, change default credentials, and replace end-of-life systems. Just as important, keep an independent temperature alarm and a local response plan.

The freezer story is dramatic. The durable lesson is not. If a networked system controls a physical process, recovery has to protect the process, not merely restore the dashboard.

Reliable automation

Automation is wonderful until it quietly stops working three Tuesdays ago. Raymond Tec builds integrations with logging, monitoring, and failure handling in mind so the boring work stays automated without becoming mysterious.

Amazon’s Ad Auction Case Is About the Price Below the Bid

The Federal Trade Commission and 22 states unsealed new details in their lawsuit over Amazon’s sponsored-ad auctions. The allegation isn’t that Amazon charged advertisers more than the maximum bid they entered. It is that Amazon secretly changed the amount the winner had to pay while describing the auction differently.

Amazon describes Sponsored Products as a second-price auction. In the simplified version, the winner pays roughly enough to beat the next qualifying bidder, not the full maximum bid. The government alleges Amazon used internal “soft reserves” after seeing the bids to push the winning price higher. The newly public complaint says costs rose by as much as 50 percent on ordinary days and 250 percent during major shopping events, with an average cost-per-click increase of about 20 percent in one 2021 test. Those are allegations. A court hasn’t decided the case.

Amazon disputes the lawsuit. It says advertisers never paid more than their bids, reserve prices are normal, and relevance controls improved sales and return on ad spending.

I don’t think “you stayed under your maximum” answers whether the pricing mechanism was disclosed honestly. A ceiling isn’t permission to invent any price beneath it. Small sellers shouldn’t assume a refund is coming, either. Preserve campaign reports, compare holiday cost per click and return on ad spend with ordinary periods, and keep the allegation separate from the evidence a court eventually accepts.

Ecommerce development

A good ecommerce site has to work in front of the homepage and behind it. Raymond Tec designs and builds ecommerce sites on Shopify, WooCommerce, Magento / Adobe Commerce, BigCommerce, and other major platforms — including the storefront, product catalogs, inventory, navigation, integrations, fulfillment workflows, and the machinery customers hopefully never have to think about.

Something Good: A Security Guide for the Business With One Employee

NIST published Small Business Cybersecurity: Non-Employer Firms, a plain-language guide for sole proprietors, freelancers, and other businesses where the owner is the only paid employee. That audience matters. Security advice often assumes there is an IT department, a compliance team, and somebody whose calendar doesn’t already contain sales, billing, support, and the work customers actually pay for.

The guide adapts the NIST Cybersecurity Framework into practical steps: inventory the devices, accounts, software, and data the business depends on; replace default passwords; keep software updated; maintain backups; limit access; and decide what happens when a system fails. It includes editable worksheets and hypothetical examples rather than stopping at “follow best practices,” a phrase that has ended many conversations without beginning any work.

If you work alone, start with the inventory. Write down the laptop, phone, email account, domain registrar, website host, payment service, cloud storage, bookkeeping system, and backup location. Add who can access each one and how you would recover it. You can’t remove access you forgot was granted, and you can’t restore data you never decided to protect.

That is not glamorous. Good. Glamour is a poor recovery control.

What to Watch Next Week

The Arkéa investigation still needs a technical explanation and the identity of the service provider if authorities can responsibly disclose it. OPSSHIELD’s cPGuard investigation is still developing, so administrators should watch for changed remediation or rebuild guidance. Amazon’s case is at the complaint stage. WorkOS has published a useful root-cause account, but the larger question is whether other companies will now look for old internal applications before an agent does it for them.

Also watch the federal AI-testing office, CAISSI. It lacks a permanent director, has stopped publishing research, and faces questions about its reporting structure. There’s nothing for an ordinary user to change today. The useful signals will be stable leadership, a clear chain of accountability, and public evaluation results that let people judge more than a company’s promise to test itself.

The practical work doesn’t need to wait. List the outside accounts that can move money or change production. Remove access that no longer has an owner. Require a second channel for dangerous changes. Separate at least one recovery copy from the identities that run the live system. And when a patch fixes an actively exploited path, verify the version, then ask what happened before the door closed.

Still in a reading mood? The Raymond Tec News archive covers security, AI, small-business technology, policy, and the places technology collides with ordinary life — without requiring a computer-science degree to get through it.

Sources / Further Reading

Trusted access, the Arkéa bank attack, and cPGuard

SonicWall, QNAP, and incident response

AI agents, permissions, and cloud costs

Refrigeration systems and physical recovery

Amazon advertising and small-business cybersecurity

AI evaluation oversight

Photo by Kindel Media on Pexels.

Related Posts

Leave a Reply

Only people in my network can comment.