Anthropic's Security Chief Says Hold On for 18 Months. Here Is What to Ask Your Board For.

On October 22, Anthropic's deputy CISO, Jason Clinton, told security leaders that the next stretch will be worse than the four days of attacks and emergency patches that came before it, and that defenders need to hold on for roughly 18 months. His advice was not to buy AI defenses. It was to go to the board and ask for help with the basics. This piece turns that advice into a specific, time-boxed request a mid-market board can approve.

What he actually said

Clinton posted on X that "as bad as things have gotten in the past 4 days, they are going to get a lot worse before they get better," and urged security leaders to use the moment to ask their boards for help with the basics, not AI, adding that "we need to hold on for ~18 months." The posts were reported by India Today and OfficeChai, among others.

He had flagged this in August, warning of a "vulnerability tsunami" with too little time to patch critical infrastructure. Asked what changes after 18 months, he pointed to what he calls "vulnzero": AI that finds and removes bugs cheaply before code ships. Until that is built into normal development, attackers get to use today's models against yesterday's code.

That is the part worth sitting with. Most of the software you run was written before any of this. It is not getting rewritten in the next year and a half.

Why the basics, and why now

October showed what the window looks like. Emergency patches went out for two critical PHP flaws, and Cloudflare's CEO said exploit attempts went from thousands to millions within hours of disclosure. The FBI told organizations to patch immediately. Around the same time, a cloud hosting platform told customers to rotate secrets after a breach.

None of those incidents needed exotic defense. They punished three ordinary gaps: not knowing what was exposed, not being able to patch it fast, and not knowing which credentials lived where. An AI security product does not close those gaps. It sits on top of them.

We made a related argument in October about who is allowed to act when an exploit lands in hours, and about the companies sitting below the AI cybersecurity poverty line. Clinton's framing adds something useful for the board conversation: a deadline. Eighteen months is a budget cycle and a half. Boards can work with that.

The board ask: five items, each with a test

Keep the request short and checkable. For each item, ask for something concrete and agree on a test you will report against every quarter until the window closes.

1. A list of everything that faces the internet, with an owner

Web servers, VPNs, firewalls, remote access tools, SaaS admin portals, and anything a vendor runs on your behalf. Include what your hosting provider and MSP manage for you.

The test: when the next critical flaw is announced, can you say within one hour whether you run the affected software, where, and who will patch it?

2. Authority and capacity to patch exposed systems in days

Ask for a standing rule that critical flaws on internet-facing systems can be patched outside the normal change window, plus the staff or partner hours to actually do it. Speed is mostly a permission problem, then a people problem.

The test: how many days did the PHP patches take you, from disclosure to done? Report that number for every critical flaw from here on.

3. Strong sign-in on every admin and remote path

Phishing-resistant MFA for administrators, remote access, email, and cloud consoles. Remove accounts nobody uses, and rotate the secrets and API keys that live in build tools and hosting platforms.

The test: name every account that can change production, and show that each one requires MFA. If a provider tells you to rotate secrets tomorrow, can you list them?

4. Backups you have restored from

Copies attackers cannot reach or alter, for the systems the business cannot run without.

The test: the date of your last real restore, and how long it took. A backup nobody has restored is a hope, not a control.

5. Logs that can answer "were we hit?"

Central, retained logs for internet-facing systems and sign-ins, kept at least 90 days.

The test: for the PHP flaws, could you show whether anyone tried them against you, and whether anything got through? Our first 24 hours playbook for edge appliance compromises covers what to look for.

How to put it to the board

Directors respond to a bounded problem with a clear ask. Something like this works:

For the next 18 months, attackers will find and use known flaws faster than we patch them. We are asking for three things: authority to patch exposed systems outside the normal change process, funding for the people or partner hours to do it, and a quarterly report against five tests. We are not asking for new AI tools.

Then report the five tests every quarter, in the same format, so the board can see the trend. Our 12 questions every board should ask make a good companion for directors who want to probe the answers.

What not to do

Do not wait for vulnzero. If Clinton is right, the tools that make bugs cheap to remove will arrive, but your existing systems will still need patching, inventory, and recovery until they are retired. And do not let the moment turn into a shopping list. The organizations that come through this window well will be the ones that can answer the five tests above, not the ones with the most products.

If you need help building the case for your board, or running the program behind it, that is what our virtual CISO services do.