Skip to content

The Nvidia Story Is a Warning About Who You Trust

· · 5 min read
The Nvidia Story Is a Warning About Who You Trust

The Nvidia story raises a personal security question. Who can reach your accounts, your devices and the AI Hires you trust with your work?

Nvidia just lost the man who helped build it.

On November 12, Jensen Huang resigned as CEO. His departure followed reports of an affair with Alice Huang, a businesswoman suspected of working for Chinese intelligence.

Why would China want access to him?

Nvidia makes advanced chips used to build and run AI. US rules limit which of those chips can be shipped to China.

Bloomberg, citing New York Times reporting, described a suspected effort to learn about US AI chips and possibly persuade Nvidia to make its most advanced chips available in China.

Her exact role remains unclear, and China denies the claims. Nvidia opened an investigation. Its board initially said it had no reason to believe company secrets or national security had been compromised. We do not yet know what information, if any, was shared.

The risk I see in this story goes well beyond Nvidia.

Why break through security if you can convince someone to open the door?

KB


The target can be you, not your computer

That is the idea behind social engineering: tricking a person into sharing information or granting access.

The bait might be a romantic interest. A business deal. A fake support agent who says your account is in danger. Someone posing as your boss and asking for one quick favor.

They give you a reason to want to help, or a reason to be afraid not to.

Then comes the request.

Send this file. Open this link. Share that code. Approve this payment.

You do not need to run a chip company to have something worth stealing. Your email, customer list, private photos or crypto wallet may be enough.

Keep the tools. Know their limits.

A YubiKey is still worth using. With its phishing-resistant login method, called FIDO2, it can stop a fake website from signing you into the real account.

But it cannot stop you from signing in yourself and sending private files to someone who fooled you.

A hardware wallet protects your private keys. If you share its recovery words, someone can take control without touching the device. Those words are the backup that restores your wallet. They are not a support code.

You can also lose funds by approving a harmful transaction. The wallet does not know whether the person asking you to send money is honest.

Strong tools reduce risk. They do not make every request safe.

Your AI Hire is part of this now

I build AI Hires at cYpher.camp because I want people to hand off real work. Not spend all day asking a chatbot what to do next.

Asking AI how to follow up with a customer still leaves you with the follow-up. Giving your Hire responsibility for it means setting the rules, connecting the accounts and deciding when it needs to come back to you.

Our AI Hire use-case walkthroughs show what those assignments can look like:

  • Keep your website current. Make the updates you approve, publish new pages and check that the changes are live. You decide what belongs on the site without doing every edit yourself.
  • Keep sales conversations moving. Ask new leads your intake questions, track their answers and put ready leads in your proposal queue. Follow up with customers who agreed to contact. Bring you the discount request or unusual job that needs your judgment.
  • Keep confirmed orders organized. Match an order to its approved quote and artwork, prepare the job sheet and send the approved proof message. Flag missing files or changes before they hold up the work.
  • Keep approved content going out. Publish approved videos, prepare subscriber emails and use the current offer and link. Work from the schedule and rules you set instead of waiting for you to remember every post.
  • Prepare the paperwork for paid work. Draft work orders from signed contracts and bids from approved scope and prices. Bring back conflicting drawings, missing terms and final bids that need your approval.

These are scripted examples, not records of live customer work. They require connected accounts, setup, permissions and AI credits.

The point is to give your Hire a clear responsibility and let it keep working within the limits you set. It brings back a result, a decision or a blocker. You should not have to direct every click.

I want that kind of help. But I also have to take seriously what I am giving it access to: customer records, business files, a website and the channels where people hear from me.

The more power you give your AI Hires, the more dangerous they become in the wrong hands.

The permission that lets your Hire publish your approved email could also be used to send a scam to your subscribers. Website access could be used to replace a real payment link. File access could expose private customer information.

The risk depends on the access you grant and the controls around it.

If you direct your Hire through Telegram or Slack, someone using your logged-in account may be able to give orders that appear to come from you. They may be able to act through your Hire without stealing every connected account's password, because your Hire already has permission to work there.

Your Telegram or Slack account can now direct work in those connected accounts.

Your phone may be locked. What about the Slack tab on your office computer? The Telegram desktop app on a shared laptop? The old tablet you never signed out of?

Check every way in

  1. Lock the devices.

Who can unlock your phone or use your computer? Use a strong passcode, turn on automatic screen lock, and lock the screen when you walk away. Keep apps and devices updated. Do not leave work accounts open on shared computers.

  1. Protect the logins.

Turn on Telegram's Two-Step Verification and Slack's two-factor authentication. If Slack uses your company login, check its multi-factor authentication, or MFA, too.

Protect the account used to manage your AI Hires and the email used to recover access. Never share login codes or approve a sign-in you did not start.

MFA adds a check when you sign in. It does not lock an account that is already open.

  1. Check where you are still signed in.

A session is a browser or device where your account is logged in.

In Telegram, open Settings → Devices and end old or unfamiliar sessions. In Slack, review Access Logs in your account settings. If something looks wrong, use Sign out all other sessions.

Check regularly, and after losing, selling or replacing a device.

  1. Limit who can give orders and what those orders can do.

Review which people and groups can direct your AI Hire. Being in the same chat should not automatically give someone control. Remove access when someone leaves your team.

Give each Hire only the access its job needs. Reading a calendar does not require permission to delete events. Drafting an email does not require permission to send it.

Use actual account and app permissions to enforce those limits. Do not rely only on telling the AI to be careful.

  1. Put high-risk actions behind a separate check.

For payments, new account access or major deletions, keep a separate, protected approval step. If your setup cannot enforce that, keep the action manual.

A second “yes” in a stolen chat account is still coming from the stolen account.

Confirm sensitive requests through a number or channel you already know, not the contact details in the request. Know how to pause your Hire and cut off its connected access if something goes wrong.

Before giving your AI Hire another job, check who else could give it orders.

Agentic & distributed systems, DeFi, and the compute economics. One email a week, no fluff.

Subscribe to the newsletter →

About the author

Keenan Benning is the founder of cYpher.camp, CTO of DeFi All Odds, and a forward-deployed AI systems engineer. He builds AI systems and writes about the engineering, economics, and ownership questions behind them.

Other projects