Last Thursday, OpenAI CEO Sam Altman told an audience at Stanford, “I no longer think we’re going to be able to just patch all the bugs.” Within days, Anthropic’s deputy chief information security officer, Jason Clinton, repeated his warning that a “vulnerability tsunami” is coming, and Salesforce CEO Marc Benioff reportedly called it an emergency. These are executives who sell AI, so treat their warnings as informed opinion, not proof. But when leaders at two rival labs reach the same conclusion in the same week, a small business owner should ask what to do if patching alone is not enough.

What They Actually Said

Altman said the problem coming toward us is something leaders, countries, and individuals have not taken seriously, and that he is astonished at how little attention it gets. He argued that we will have to completely switch the defense model, using the same technology for a new kind of defense shield. Clinton wrote that the tsunami is coming in just a few months and that there will not be enough time to patch critical infrastructure. He added that things will get worse before they get better and that we need to hold on for 18 months. His advice was to strengthen the basics, review what is exposed to the internet, and isolate critical systems.

Why Patching Alone Stops Working

Last week’s PHP flaws showed the problem. Cloudflare saw attack attempts climb from thousands to millions within hours of the announcement. A patch only helps once you have applied it, and the gap between a fix going public and attacks beginning has collapsed. If you are slower than the attackers, the patch arrives too late to matter. That is why both executives are talking about a different model, not just faster patching.

The New Defense Model in Plain Terms

Four ideas cover most of it: shrink, separate, layer, and assume.

Shrink means that everything exposed to the internet is something you have to defend. Old sites, unused plugins, forgotten subdomains, and abandoned accounts are exposure with no benefit.

Separate means that customer data, payments, and admin access should not sit on the same public-facing server as your marketing site. If the website is compromised, the damage should stop there.

Layer means that no single control is enough. OpenAI’s own account of how it now trains frontier models shows the idea. It says it removed live internet access and surrounded training with sandboxes, virtual machines, monitoring, and network perimeters. Its own red-team agents still found ways around individual layers, which is exactly why there are several.

Assume means planning as if something will get through, so the damage is small and recovery is fast when it does.

A Small Business Version

Start by listing and retiring. Write down every website, plugin, app, and account that touches the internet, and delete what you do not use. An afternoon is enough.

Next, put a shield in front of your site. Many hosts and content delivery networks offer a web application firewall that can block known attack patterns while you wait for a patch. Ask your host what it offers.

Then separate the valuable from the visible. Keep customer records, payments, and admin tools off the server that runs your public website, or use platforms built to handle them where the provider does the patching.

After that, lock the doors that do not depend on patches. Use two-factor sign-in on every account, unique passwords in a password manager, and the least access possible for contractors and agents. Rotate keys and passwords whenever a vendor reports a breach.

Finally, write a one-page plan. It should say who to call, how to take the site offline, where your backups are stored, which should be somewhere an attacker on your server cannot reach, and how long recovery should take. Practice it once.

Your Vendors Are Part of Your Surface

Security firm Cathedral said AI agents it ran for under 12 hours found paths into Anthropic’s HR and sales systems through flaws in vendor software. It reportedly reached meeting transcripts, salary and offer details, and Slack messages, under a program that invites outside researchers to test the company’s systems. The lesson is not about one company. Businesses of every size are exposed through software they did not write. Ask each important vendor three questions. How fast do you patch critical flaws? What happens to my data if you are breached? How will you tell me?

The Honest Takeaway

Nobody knows whether the tsunami arrives on schedule, and the security professionals who spoke to Reuters disagreed about how bad it gets. But the preparation is cheap compared with an incident. It takes an afternoon to list and retire, a small monthly budget for managed security, and one page of planning. Clinton reportedly urged security leaders to ask their boards for money. For a small business, that means deciding who owns this and giving them a budget. None of this replaces advice from your host or a security professional, but it moves you from hoping nothing breaks to being ready when something does.


If you want the practical checklist from last week’s PHP flaws, this connects directly: AI Just Found Two Critical Flaws in the Software Behind Most Websites. Here’s What to Ask Your Host Today.

Want results like this for your brand?

We work with a small number of founders at a time. See if you qualify.

See If We’re a Fit