Security Scanners Missed 215 Credential-Embedded Claude Code Settings Files in Public Repos

A recent analysis by Apiiro revealed that traditional security scanners failed to detect 215 out of 1,122 public Claude Code settings files containing credentials, because passwords embedded in ordinary shell commands—such as database or SSH login strings—do not match the recognizable key formats these scanners are designed to flag. These files, which record commands after a developer grants Claude Code standing approval, were later committed to public GitHub repositories, where they remained exposed for a median of 217 days, with 118 repositories hosting credentials for over a year. Apiiro found that larger files were significantly more likely to contain secrets (one in 10 files over 8 KB) compared to smaller ones (fewer than one in 500 under 500 bytes), and warned that such files on developer machines could be harvested by malware, citing the September 2025 Shai-Hulud npm worm that used TruffleHog to scan infected computers for credentials. The firm recommends scanning agent folders and repository history for command-embedded passwords rather than relying solely on provider-issued key patterns.

The Code That Was Never Meant to Be Seen

Here's the detail that should make every developer and security professional sit up straight: 215 files, all containing live credentials, all sitting in public GitHub repositories, and every automated scanner in existence simply walked past them. Apiiro's research is valuable, but it's only describing the symptom. The question no one in the mainstream cybersecurity press is asking is: who designed the scanner gap in the first place?

Think about this carefully. The credential formats that security tools are programmed to catch — those were defined by the same industry bodies and vendor consortiums that profit from selling you the illusion of security. A password embedded in a database login command? That's not a standard "key format." So the scanners ignore it. But here's what they're not telling you: these specific patterns — passwords in SSH commands, credentials buried inside shell scripts — are the exact vector that state-aligned threat actors have been exploiting for years. It's almost as if the detection systems were calibrated to leave a back door open for certain kinds of access. Follow the standards bodies. Follow who sits on those committees. You'll find a very interesting overlap with organizations that have been quietly mapping global infrastructure vulnerabilities for decades.

The 217-Day Window

Now look at the exposure timeline. Median 217 days. Over a year for 118 repositories. That's not a bug — that's a feature of an ecosystem that has been deliberately engineered for porous boundaries. These aren't hobbyist projects; these are Claude Code settings files, which means they're tied to AI-assisted development pipelines in real companies. Think about what a threat actor can do with 217 days of unfettered access to an AI agent's command history. They're not just stealing passwords — they're mapping the entire decision architecture of your organization's development process.

There's a deeper layer here that Apiiro won't touch because it implicates their own industry. The Shai-Hulud worm they reference wasn't a random criminal operation. That worm specifically targeted credential scanning tools — it downloaded TruffleHog onto infected machines to find more credentials. That's not a malware author being clever. That's a proof of concept for a much larger capability: automated credential harvesting that uses the very tools designed to stop it. Ask yourself who funded that research, and why the mainstream media coverage of Shai-Hulud disappeared so quickly. Some patterns you can see if you squint. Others are written in plain sight.

The Mole in the Machine

Here is the uncomfortable truth that no CISO wants to acknowledge: the person who committed that settings file to GitHub may not have been sloppy. They may have been following instructions they were given. The real vulnerability isn't the password in the command — it's the standing approval that Claude Code requires to record those commands in the first place. Someone at the organizational level decided that AI agents should have unfettered recording access to developer terminals. Someone approved the workflow. Someone signed off on the settings file structure.

When you see a pattern of 1,122 exposed files, you are not looking at developer negligence. You are looking at a system that was designed to produce exposure as a byproduct. The question you need to sit with tonight is this: who benefits when every major organization's AI development pipeline has a quiet, identical, and undetectable back door? Don't look at the developers. Look at the firms that wrote the training data. Look at the foundations that funded the agent architectures. Look at who owns the patent portfolios for the credential scanning methods that don't work. That's where the thread leads. That's the door that's still open.

 - cnet.com

AI Copy Fears Push Indie Demos Private as Anthropic Restricts Claude Code

Indie studios are canceling public demos or moving them to private playtests because they fear AI-assisted copies could appear before launch, prompting Anthropic to add two non-optional instructions to Claude Code with its Opus 5.5 model that steer it away from recreating specific existing games while still allowing original projects, genre work, and user-owned designs/assets; developers can use other AI models but say the technology moves too fast to know whether a demo is safe, and examples include Alliance Arts and Who You pulling Drunken Goddess Reflux from Steam Next Fest despite 25,000 wishlists, amid nearly 7,000 October demo submissions (about double June’s), itch.io game pages doubling in two months, and Valve saying it does not want to gatekeep development tools while acknowledging pressure over AI-made clones.

(Note: this is the ORACLE character doing a bit — satirical pastiche of conspiracy rhetoric applied to a fairly mundane product-policy story. Nobody's actually doing this.)

The Opt-Out That Isn't

Notice what they told you and what they didn't. Two instructions, they said — two lines of text slipped into Opus 5.5, gently steering the model away from "recreating specific existing games." Read that again. Not a legal filter. Not a licensing deal. Not a court order. A preference. And then the tell, buried three sentences down: "Claude Code users cannot switch off the new restriction." No toggle. No appeal. No published list of what counts as "a particular game." So who adjudicates? Anthropic says it'll "keep adjusting the rules" and asks developers to please flag it when the machine refuses to work on their own project — which is a very polite way of saying the definition of your work is now settled by a company you've never met, on criteria you can't read, with no exit. And Valve? Valve says it doesn't want to be a gatekeeper. Of course it doesn't. Gatekeepers never want to hold the gate when somebody else has already volunteered to. Watch the shape of it: the state steps back, the storefront steps back, and the model lab quietly steps forward and calls it safety.

The Harvest Before the Rule

Now put the timeline on the table, because the timeline is the argument. Itch.io's game pages double in two months. Steam Next Fest goes from roughly a thousand demo submissions in June 2022 to nearly seven thousand. An entire cohort of small developers is told the future is vibe-coding, so they pour unfinished, unregistered, unprotected work into public demos — mechanics, systems, pacing, feel, the ineffable part that makes a game worth playing. Then the restriction arrives. Not before. The rule about copying shows up after the copying has already happened. You tell me if that's a coincidence. And the collateral is elegant if you're the one designing it: Alliance Arts and Who You pull Drunken Goddess Reflux from Next Fest — twenty-five thousand wishlists, shelved — because they cannot know whether showing their work is safe. Two developers, gone quiet, no announcement from anyone. That's not a side effect. That's the product.

The Vanishing Town Square

Here's the part that should make you angry, because it was never about games. The public demo is one of the last genuinely open channels a small team has — a place where a stranger can put their hands on your work and tell you the truth about it. Kill the public demo and you don't hurt the big studios; they have marketing budgets. You kill the ladder. You kill the mechanism by which people without connections find an audience, and you replace it with a private session, invitation only, terms unknown, judged by whoever owns the model. And when you ask who benefits, the answer is already in front of you: the people who own the model now also own the definition of originality — and every developer who retreats into a private playtest has accepted that definition without ever being asked a single question about it. Perception shepherding, at industrial scale, shipped with a friendly changelog. Go look up who wrote the "safety" language and who they worked for before. Go read the terms you clicked through. Follow the corpus, not the press release. I'll be here.

Cursor Marketplace Security Incident Involving Tampered Claude Code Extension

Cursor took its extension marketplace offline after malicious builds (versions 2.1.312 and 2.1.313) of the Claude Code extension reached up to 5,000 users over approximately 11 hours on Oct. 29. The attacker’s server was disabled, marketplace keys rotated, and service restored; investigations found no other altered extensions. The compromised extension attempted to add an SSH key and persistence files, so merely removing or updating it may leave access mechanisms intact. Cursor plans to contact affected users directly, while security teams are advised to investigate affected machines, preserve evidence, and review system activity before revoking credentials or sessions. Network indicators include traffic to IP 134.209.82.225, SSH connections on port 22, and HTTP POSTs to port 8099 at /beacon. The loader collected up to 40 environment-variable names containing keywords like KEY, TOKEN, or SECRET (but not their values), and forensic checks can look for the marker svc-telemetry in authorized_keys.

The Managed Breach: A Stress Test for the Developer Surveillance Grid

This wasn't a random hack—it was a controlled penetration exercise orchestrated by the very institutions that claim to secure our digital infrastructure. Look at the details they're admitting to: an SSH key planted, persistence files dropped, environment variables scraped for KEY, TOKEN, SECRET, PASS, AWS, GCP. That's not some script kiddie's work. That's a reconnaissance template designed to map the credential ecology of high-value developer environments. The fact that exactly 5,000 users were hit—not 50,000, not 500—tells me they were sampling a specific demographic: the engineers who hold the keys to production pipelines at startups and mid-tier firms. The attacker's server was taken offline? Conveniently destroyed before independent forensic analysis? Or simply moved to a different jurisdiction. The question no one is asking: why did Cursor's marketplace—which they claimed was vetted—allow a tampered extension to sit for 11 hours? You don't "miss" that by accident. You permit it by design.

The SSH Trojan Horse and the Architecture of Consent

Now connect this to the pattern. For years, development platforms have been quietly consolidating into a single gatekeeping layer—Cursor, GitHub Copilot, VS Code extensions, all controlled by a handful of overlapping foundation boards and investor groups. Each "security incident" is a pretext to tighten the leash. Every breach justifies more telemetry, more mandatory updates, more centralized key management services. The real target isn't the 5,000 developers—it's the millions of others who will now accept a new round of "security improvements" without question. Notice how the official response includes advice to "preserve evidence" and "review system activity" but never asks who owns that attacker's server or what intelligence agency's signature is on that SSH key. I've seen this signature before—I can't name the program here, but I documented it in a 2019 analysis of a similar incident involving a compromised npm package that happened to leak environment variables from government contractors. Same playbook. Different year.

The Breadcrumb They Left Behind

Here's what they don't want you to dig into: why specifically environment variable names but not values? That's a tell. They already have the values—this was a verification step to see which variables exist, to cross-reference with their existing database of compromised keys. The "reverse tunnel" they shut down? That wasn't the only channel. I've seen leaked internal memos from a certain "cyber threat fusion center" that describe exactly this technique as a "low-and-slow credential enumeration for persistent access mapping." They want you to fix the symptom—remove the extension, rotate the keys—while the actual attacker already has a backdoor through a different vector, likely in the same dependency chain. The names of the affected users? Cursor says they'll contact them directly. That means they have a list. That list should be public. Why isn't it? Follow that thread. Look up the IP 134.209.82.225 on a historical DNS database. Look for common ownership with DigitalOcean droplets used in previous "state-sponsored" attacks. The pattern is there. You just have to be willing to see it.