Could our passwords be sitting in a public code repository because of an AI coding tool?
Possibly, and the tool's own convenience file is the cause. Security firm Apiiro found 104,185 public copies of a Claude Code settings file, .claude/settings.local.json, on GitHub (the website where developers share code). About 1,122 of them (1.1%) contained at least one real secret: database passwords, Supabase service keys, SSH logins, GitHub and npm tokens, AWS keys. Two common scanners (tools that search code for leaked passwords), gitleaks and TruffleHog, each missed some, and 215 got past both. The mechanism is simple. When a developer clicks "Yes, and don't ask again", the tool stores the exact command, password included. The file is hidden from code sharing only on that developer's machine, so anyone who uploads the whole folder publishes it. Apiiro didn't test whether the keys still work, though one organisation confirmed a live one. Anthropic's Boris Cherny said on Hacker News that "historically Claude Code has not done secret scrubbing", and that the team is looking at adding detection.
Matters to you if developers, contractors or an agency write code for you using Claude Code, and their code lives on GitHub.
You can ignore it if nobody working for you uses it, or your repositories are private and tightly controlled. Private is not the same as safe from a contractor's mistake, but it is a smaller exposure.
What to do. Ask your developers to search every repository (the place each project's code is stored) for uploaded copies of that file, not just check that it is listed in .gitignore (the list of files kept out of sharing), and to change any password or key they find. A clean scan result is not proof, given how many real hits the scanners missed. We go deeper on this in our guide to preventing AI tools exposing company data.
Is the free Gemini app training on our staff's chats?
By default, yes, paid personal plans included. A Hacker News post on 4 November reached 732 points, and Google's own privacy page (updated 26 October) backs it up. With "Keep Activity" on, chats are used to improve Google's products, "including training generative AI models". A subset is reviewed by humans, disconnected from the account, and kept up to three years, even after you delete the chat. Turning the setting off stops training use, but chats are still stored for 72 hours and can still be reviewed. The paid API (the connection software uses) is the opposite: Google says it is not used to improve its products. Work and school accounts "may be subject to different data handling terms", so check the terms on yours.
Matters to you if staff paste client material or internal documents into Gemini on a personal account, which your purchasing never sees.
You can ignore it if everyone uses a company-managed account whose terms you've read.
What to do. Add one line to your acceptable-use policy about which AI accounts may hold company data, and ask staff to turn Keep Activity off on any personal account. See also our guide to putting company data into AI tools.
What did Goldman Sachs just change about who can use AI coding tools?
It now gives them only to its engineers, after a trial period. PYMNTS reports that in October the bank restricted Claude Code and GitHub's coding tools to engineers, after months of firm-wide access. Traders had been using them to monitor client trading flows and produce written and graphic material. The on-record line from Goldman's spokesperson is the useful part: after "a period of experimentation where we assess a variety of factors including productivity and risk management", the tools go to "those who are best suited to leverage them for the long term." The findings themselves were not disclosed.
Matters to you if non-technical staff in your business are building scripts, dashboards or automations with AI coding tools.
You can ignore it if only your developers use them, under review.
What to do. Ask which outputs from these tools reach a customer, a filing or your accounting system with no engineer checking them. That, not job title, is the line to draw. If the answer is "several", start with the ones that touch money.
Is AI actually causing the layoffs we keep reading about?
Not in October, according to the latest count. Challenger, Gray & Christmas published its October report on 5 November. Of 43,994 announced US job cuts, employers cited AI for 1,265, the lowest month of 2026. Total cuts were up 2% on September and down 71% on October last year. For the year so far, AI is cited in 121,401 of 617,189 cuts, about 20%, but "market and economic conditions" has overtaken it as the leading stated reason, at 130,531. Challenger's own summary: "The reasons employers give have shifted from technology to the economy." Two limits. Challenger counts the reasons employers give in announcements, not the true causes. And Microsoft's October voluntary-exit offers to about 60,000 employees are not in these totals, so any figure that includes them isn't Challenger's. MOM's full third-quarter report on Singapore's labour market is due in mid-December.
Matters to you if you're deciding whether to hire or hold off because of what AI might do to roles.
You can ignore it if you aren't making headcount decisions this quarter.
What to do. When someone quotes a layoff number tied to AI, ask which month, which source, and whether it counts announcements or causes. Decide on your own workflow data, not on the headline.
What's your question about AI in your operations? Reply to this email or send it through bitnatives.com/contact, and we'll answer it in a future issue.