The AI Vulnerability Tsunami: What Anthropic and Bill Gates Are Warning Businesses About

Last week a client asked me whether they should be worried about "the AI hacking thing" they kept seeing in the news. I told them the truth: yes, a little. Not panicked, but worried enough to fix the boring stuff now instead of next spring.

Two warnings got a lot of attention this month, and they come from people who don't usually exaggerate. One came from Jason Clinton, Deputy Chief Information Security Officer at Anthropic, the company behind Claude. The other came from Bill Gates. They describe the same problem from two sides, and if you run a business in Canada, both are worth five minutes of your time.

The short version

• Anthropic's Jason Clinton warned in August 2026 that a "vulnerability tsunami" was coming, with more software flaws found than anyone can patch in time. On October 22 he said things will "get a lot worse before they get better" and that we need to "hold on for ~18 months."
• Bill Gates said on October 11 that cyberattacks are the most immediate AI risk, driven by "bad people" using AI models with the safeguards removed, and predicted "a lot of negative surprises" over the next year.
• What to do: both point to the same answer. Get the basics right: patching, multi-factor authentication, backups you've actually tested, and limiting what's exposed to the internet.

What did Anthropic's security chief actually say?

On August 20 or 21, 2026 (reports differ by a day), Clinton posted on X: "The vulnerability tsunami is coming in just a few months and there won't be enough time to patch critical infrastructure." At the time, not many people paid attention.

Then October happened. Researchers using frontier AI models found two critical remote code execution flaws in PHP, the language behind a huge share of the world's websites. The flaws, nicknamed PHP2Shell, had been sitting in the code for more than a decade. Government portals, hospital sites and school services went offline as a precaution while operators patched. Some people started calling it "Black October."

On October 22, Clinton followed up:

"As bad as things have gotten in the past 4 days, they are going to get a lot worse before they get better. Security leaders, never let a crisis go to waste: please go to your boards and ask for help right now. Not for AI, for the basics. (We need to hold on for ~18 months.)"

Two things stand out to me. First, the guy who works at an AI company told people not to spend their emergency budget on AI tools. Second, he's actually optimistic about the long term. He talks about "vulnzero," a future where small, cheap AI models catch security bugs before software ever ships. The pain is the gap between now and then.

To be fair, the 18-month figure is one person's estimate in a personal post, not an official Anthropic statement. But the incidents behind it are real.

What did Bill Gates say about AI and cybercrime?

In interviews on October 11, Gates was blunt. The danger right now isn't AI taking over. It's people.

"The real danger right now is not the AIs getting out of control. We need to work on that, but that's years away... It's really bad people using AI, particularly in AI where you can take out all the safeguards."

He said the area we'll see "most dramatically in the next year is cyberattacks," and that when you combine "bad people, no safeguards, and then the latest models," we're going to have "a lot of negative surprises." He also made a point every business owner already knows from experience: "Extortion is a tried and true business model." Ransomware isn't going anywhere. AI just makes it cheaper to run.

A couple of weeks later, talking to The Observer about the PHP bugs, he went further: "You know there'll be thousands of things like that because the way that we wrote the software wasn't architected richly enough."

Why this matters for small and mid-sized businesses

I hear "we're too small to be a target" a lot. Here's the problem with that thinking in 2026. Attackers don't pick targets one by one anymore. They scan the whole internet for a known flaw and hit whatever's unpatched. AI makes that scanning, and the phishing emails that follow, faster and more convincing.

So the risk isn't that someone decides to come after your accounting firm or clinic. It's that your old VPN box, an unpatched WordPress plugin or a forgotten test server shows up on a list. Then you're in the queue with everyone else.

If you're in health care in Ontario, there's another layer. A breach of patient information means PHIPA notification obligations. I wrote about that recently in PHIPA changes in 2026.

What should businesses do over the next 18 months?

None of this is glamorous. That's sort of the point. Here's what I'm telling our own clients:

1. Patch faster than you used to. Critical updates on anything facing the internet should go on within days, not at the end of the month.
2. Know what you expose. Make a list of every system reachable from the internet: firewalls, VPNs, remote desktop, websites, cloud apps. Close or hide anything that doesn't need to be public.
3. Turn on multi-factor authentication everywhere, especially email, remote access and admin accounts.
4. Test your backups. Not "we have backups." Actually restore something. Keep at least one copy offline or immutable so ransomware can't reach it.
5. Isolate what matters most. Clinton specifically urged operators to separate critical systems from outside networks. For a small business that means segmenting your network so one infected laptop can't reach everything.
6. Train people for AI-written phishing. The spelling mistakes are gone. Teach staff to verify any request for money, passwords or urgent changes through a second channel.
7. Have a plan for a bad day. Who do you call, how do you keep working and who tells customers? Write it down before you need it.
8. Ask for budget now. Clinton's advice works for small businesses too. If you've been putting off a firewall replacement or better monitoring, this is the moment to make the case.

Frequently asked questions

What is the "vulnerability tsunami"?

It's Jason Clinton's term for the flood of software vulnerabilities that AI models are now finding faster than organizations can patch them. He first used it in August 2026 and repeated the warning on October 22, 2026.

Who is Jason Clinton?

Jason Clinton is the Deputy Chief Information Security Officer at Anthropic, the AI company that makes Claude. His warnings were posted on his personal X account.

How long will the vulnerability tsunami last?

Clinton estimated about 18 months, after which he expects cheap AI tools to help developers find and fix bugs before release. Cloudflare's CEO Matthew Prince has made a similar forecast of "a couple years" of difficult weeks.

What did Bill Gates say about AI cyberattacks?

Gates said in October 2026 that cyberattacks are the most immediate AI threat, caused by criminals using AI models with the safeguards removed, and predicted "a lot of negative surprises" over the next year.

Do small businesses need AI security tools to stay safe?

Not first. Both warnings point to the basics: fast patching, MFA, tested backups, network segmentation and staff training. Those stop most attacks, AI-assisted or not.

A final thought

I've been doing this long enough to have seen a few "the sky is falling" moments. Most of them came and went. This one feels different because the warnings are coming from inside the AI industry itself, and because October already proved the first part right.

The good news is that the fix is mostly things you already know you should be doing. If you'd like a second opinion on where you stand, book a free 30-minute IT health check with the OITC team or get in touch. You can also read about our cybersecurity and threat protection services and disaster recovery.

Sources

• Daily Mail: Internet faces security crisis as Anthropic chief warns it will "get a lot worse" (Oct 23, 2026)
• India Today: Anthropic security chief says hold on for 18 months (Oct 23, 2026)
• Aikido: After PHP2Shell, it gets worse, then defenders win (Oct 29, 2026)
• France 24 / AFP: Bill Gates warns cyberattacks most imminent AI threat (Oct 11, 2026)
• Economic Times: Bill Gates says AI's immediate danger comes from people (Oct 12, 2026)
• The Observer: Bill Gates warns AI has exposed the internet's soft underbelly (Oct 25, 2026)