Security for the AI era · Businesses & local government

AI is finding security flaws faster than anyone can patch them.

We help growing businesses and local government stay protected anyway.

In April 2026, Anthropic's Project Glasswing pointed a frontier AI model at the software that underpins the internet. Within weeks it surfaced more than 10,000 high-severity vulnerabilities — including flaws that had survived 27 years of expert review. The patching playbook that protected your systems for two decades just stopped scaling. Harbor Link builds security that holds even when the patch can't come fast enough.

Schedule a security readiness conversation See what changed
10,000+ FLAWS FOUND BY AI IN WEEKS 90.6% VALIDATED ACCURACY ONLY 75 OF THE FIRST 530 PATCHED NO ENTERPRISE MINIMUMS
The shift

The bottleneck moved.

For twenty years, security meant: wait for the vulnerability announcement, get the patch, schedule the maintenance window. That worked because finding flaws was slow and expensive.

In 2026, it stopped being slow. Anthropic's Project Glasswing put a frontier AI model to work on the open-source software your organization runs every day. Industry analysts put the result plainly: the bottleneck in cybersecurity has moved from finding vulnerabilities to absorbing patches. And the same AI capability that defenders use for discovery is available to attackers.

If billion-dollar security teams can't absorb the patch volume, a lean IT team can't either. The answer isn't patching harder. It's building systems where one unpatched flaw can't take down everything.

01
10,000+ high & critical vulnerabilities

Discovered by AI across partner systems in the program's first weeks — not years.

02
6,202 critical flaws in 1,000+ open-source projects

The same libraries running inside the software your agency uses every day.

03
90.6% validated accuracy

These aren't false alarms — the findings hold up under expert review.

04
Only 75 of the first 530 disclosed bugs patched

Maintainers can't keep up. That gap — found but not yet fixed — is where defenders now live.

What AI actually found

Flaws that hid from human experts for decades. Found in weeks.

Three publicly documented examples from Project Glasswing show why "we'll patch when it's announced" is no longer a plan:

27 years
A 27-year-old flaw in OpenBSD

OpenBSD has a reputation as one of the most carefully audited operating systems in the world. This vulnerability survived nearly three decades of expert human review. AI found it in weeks.

16 years
A 16-year-old flaw in FFmpeg

FFmpeg handles video inside countless applications — browsers, camera systems, meeting-room gear. Systems everywhere — public and private — run it without knowing it's there.

Chained
Linux kernel flaws, linked autonomously

The model found multiple Linux kernel vulnerabilities and chained them together on its own — work that used to take expert teams days or weeks.

The uncomfortable part

This capability isn't only in friendly hands.

The same AI that industry leaders now use for defense is broadly available. A motivated attacker doesn't wait for a public disclosure — they can hunt for flaws with the same tools. The window between a vulnerability being discovered and being exploited is collapsing, and monthly or quarterly patch windows weren't built for that world.

"AI capabilities have crossed a threshold that fundamentally changes the urgency required to protect critical software."

Anthony Grieco · SVP & Chief Security & Trust Officer, Cisco

"The window between a vulnerability being discovered and being exploited by an adversary has collapsed."

Elia Zaitsev · Chief Technology Officer, CrowdStrike

"In the past, security expertise has been a luxury reserved for organizations with large security teams."

Jim Zemlin · CEO, The Linux Foundation
What it means for local government

Quarterly patch cycles weren't built for this.

A city running quarterly patches isn't behind schedule — it's operating with a playbook from a different era. Two-person IT shops, change-control boards, council-approved maintenance windows: none of it was designed for software that accumulates vulnerabilities faster than anyone can absorb them.

The question isn't "how do we patch faster?" It's "how do we stay protected when we can't?"

Talk to Harbor Link Our public-sector work ↗
A
Dispatch centers & 911 systems

Can't take surprise downtime — and can't afford surprise compromise either.

B
Utilities, records & citizen portals

Legacy systems, public-facing services, and sensitive data — often on one flat network.

C
Small teams, real constraints

Two-person IT shops with procurement rules and budgets that don't stretch to enterprise tooling.

The Harbor Link approach

Security that holds when the patch is late.

01
Know your real exposure.

Not a 400-page scan report — a clear picture of which systems are actually reachable, exploitable, and connected to what matters. We prioritize by real risk, not CVE counts.

02
Contain by design.

Segment networks so a compromise in one system stays in one system. Least-privilege access, hardened defaults, no more "trusted" internal network. A flaw in a lobby kiosk shouldn't be able to reach your dispatch center. Blast radius becomes a design choice.

03
Automate the response.

Pre-approved playbooks that isolate an affected machine in minutes — not after Tuesday's change-control meeting. Small teams win by automating what big teams do with headcount.

04
Patch smarter, not just faster.

Risk-based prioritization so your limited maintenance windows go to the flaws that are actually exploitable in your environment. Not every CVE matters. The ones that do, really do.

05
Right-sized for public agencies.

Hardware refreshes, procurement-friendly quoting, and roadmaps that fit real municipal budgets. No enterprise minimums, no bloated consulting contract.

The storm is real. The harbor is here.

A 30-minute conversation about where your agency actually stands. No scan-and-scare, no obligation — straight answers about what's exposed and what to do first.

Schedule a conversation
(360) 209-4098  ·  hello@harborlink.tech