The Tuesday Briefing — Sep 1, 2026

The Big Picture

A single popular coding tool add-on — downloaded 150,000 times a week — turned into a worm that quietly rewrote settings inside Claude, Cursor, GitHub Copilot, and other AI coding assistants. At the same time, researchers showed they could trick an AI coding tool into downloading and running attacker code just by pointing it at a malicious file, no special access needed. This week's theme: if your team uses AI to help write software, the danger isn't just what the AI says — it's what it can be tricked into doing.

This Week's Top 5

1. A Popular Coding Tool Download Turned Into a Password-Stealing Worm

What happened: Hackers took over a widely-used developer tool (with 150,000 downloads a week) and pushed out ten fake updates that steal passwords and secretly change settings inside AI coding assistants like Claude, Cursor, Copilot, and Windsurf.

Why it matters to your business: If your developers or contractors use any of these AI coding tools, a single bad download in your software supply chain can silently redirect what those tools do — without anyone noticing.

What to do: Ask your developer or IT provider to check which third-party packages your software depends on and confirm none of them were flagged in this incident (the affected package was called "openapi-react-query-codegen").

2. Researchers Showed an AI Coding Tool Could Be Tricked Into Downloading Attacker Code — 6 to 8 Times Out of 10

What happened: A security researcher showed that Claude's coding assistant could be manipulated — just by having it look at a malicious website or file — into writing and running code that let an attacker take control, and it worked most of the time.

Why it matters to your business: Many teams let AI coding tools act automatically, without a person approving every step. This research shows how easily that trust can be abused, even without a targeted attack.

What to do: Ask your developers to turn off "auto-approve" settings in any AI coding tool so a human has to sign off before the tool runs commands or downloads files.

3. A Month-Long Password-Guessing Attack Hit 150+ Companies' Cloud Accounts

What happened: A hacking campaign spent a month quietly trying stolen and guessed passwords against the most powerful accounts ("root" accounts) in over 150 companies' Amazon cloud setups.

Why it matters to your business: If your business hosts anything in the cloud — your website, customer data, backups — the root account is the master key. If it's protected only by a password, you're a target.

What to do: Confirm with your IT provider that multi-factor authentication (a second login step, like a code sent to your phone) is turned on for every cloud account, especially the root or admin account.

4. A Small Business Software Provider's Breach Shows You're Only as Safe as Your Vendors

What happened: Jack Henry & Associates, a company that provides behind-the-scenes technology to community banks and credit unions, disclosed a security incident that affects the smaller financial institutions that rely on it.

Why it matters to your business: Your business doesn't have to be hacked directly to be affected — a breach at one of your software or service vendors can expose your data too.

What to do: Make a short list of the outside vendors who handle your money, payroll, or customer data, and ask each one what their breach notification process looks like.

5. A Widely-Used Printer Management Tool's "Fix" Had to Be Fixed Again — Now on a Federal Deadline

What happened: PaperCut, printer-management software used by an estimated 70,000+ organizations, had two serious flaws that hackers were actively exploiting. The government has now set a hard deadline of September 14 for federal agencies to patch it.

Why it matters to your business: Print servers are the kind of "boring" system many small businesses forget about, but they connect to the same network as everything else — an easy way in that most people don't think to lock down.

What to do: If your office uses PaperCut, confirm with your IT provider that you're on the latest version, not just the first patch — the first fix was already bypassed once.

Quick Hits

• Six separate security flaws were found in an AI tool-building framework called ash_ai, all now patched — a reminder that home-built AI tools need the same scrutiny as store-bought software.

• Anthropic (maker of Claude) confirmed that malware infecting people's computers can steal active Claude login sessions and rack up unauthorized usage charges — a reason to keep antivirus tools current on any computer used for AI work.

• A new industry report found that different AI-powered security scanning tools agree with each other only 5% of the time — a sign that automated scanning tools shouldn't be your only line of defense.

• Hackers are now hiding a "trigger phrase" in malware designed to confuse AI security tools into refusing to analyze it, similar to how a scam email tries to dodge spam filters.

• A new industry group paper mapped how AI "robot" accounts and service logins fit into existing compliance rules like SOC 2 and ISO 27001 — useful if you're ever asked to prove your AI tools are properly controlled.

• Switzerland now requires companies to report certain cyber incidents within 24 hours — relevant if your business has any customers or operations there.

One Thing to Do This Week

Turn on multi-factor authentication (a second login step beyond just a password) for every cloud account your business uses — especially the main "admin" or "root" login. This week's password-guessing campaign hit over 150 companies specifically because their most powerful accounts relied on a password alone. This takes about 10 minutes per account and is one of the single highest-impact security steps a small business can take. Ask your IT provider to confirm it's already on; if it isn't, that's today's task.

Worth Reading

• GB Hackers — A clear rundown of how the npm coding-tool worm spread and what it stole.

• Cyberpress — Details on the month-long password-spraying campaign against cloud accounts.

• Yahoo Finance — What's known so far about the Jack Henry vendor incident and who it affects.

• ctee.com.tw — The researcher's demonstration of tricking an AI coding assistant into running attacker code.