Patch Gap
Simple Definition
The patch gap is the time a system stays exposed to a vulnerability after a fix for it already exists, because nobody has installed the fix yet.
Where It Sits in a Vulnerability’s Life
A security flaw usually moves through four steps:
- Discovery. A researcher, a vendor’s own team, an AI tool, or an attacker finds the bug.
- Disclosure. It is reported to the vendor, usually privately at first.
- Fix published. The vendor ships a patch and, often, an advisory describing the problem.
- Fix deployed. Each organization and person actually installs it.
The patch gap is the time between steps 3 and 4, and it is the step that varies most. Some systems update within hours. Others run the old version for months because updates need testing, a reboot nobody schedules, or a device nobody remembers owning. The same gap appears further down a supply chain: a browser built on another project, or a Linux distribution packaging an upstream library, can ship a fix days or weeks after the original.
Why Published Fixes Raise the Stakes
Before a fix is published, few people know the vulnerability exists. Once it is out, the patch itself points at the problem: comparing the old and new code shows what changed, a long-standing technique known as patch diffing. Not every disclosed bug is practical to exploit, and many need specific conditions. But a public fix plus an unpatched system is the situation attackers look for.
AI Is Speeding Up the Discovery Side
AI models now find vulnerabilities in volume. OpenAI says its restricted GPT-6 Cyber model produced reproducible test cases for 81 previously unknown Firefox vulnerabilities that Mozilla rated high severity, and that Mozilla has begun landing fixes. More bugs found and fixed is good for everyone. It also means more security updates, more often, and the same AI-assisted analysis that helps defenders can help an attacker understand a published fix. The defensive answer is unglamorous: make your own patch gap shorter. See cyber model for how access to these models is restricted.
Example
A browser vendor ships a security update on a Tuesday. One company’s laptops install it automatically by Wednesday. Another company pins browser versions and approves updates once a month, so its laptops run the vulnerable version for weeks while the fix and its advisory are public. The vulnerability is the same; the exposure is very different.
Shrinking It
- Automatic updates for browsers, operating systems, and common apps wherever you can allow them.
- An inventory. You cannot patch software you do not know you run, including AI agent tools, MCP servers, plugins, and their dependencies.
- Priorities. Patch internet-facing systems first, and act fastest on advisories the vendor marks as actively exploited.
- Stopgaps. When a patch cannot go in yet, turn off the affected feature or restrict network access to the system until it can.
Related Terms
- Cyber Model, AI models specialized for security work
- AI Red Teaming, deliberately probing systems for weaknesses
- AI Permission Hygiene, keeping agent access narrow so one flaw has less reach
- Agent Sandbox, limiting what an agent can touch if its tools are vulnerable
Continue learning
Explore related guides, tools, workflows, and prompts that help you go deeper into this topic.
Browse all AI terms.
Learn termSee these concepts in practice.
Open workflowA simple explanation of this AI concept.
Learn termA simple explanation of this AI concept.
Learn termA simple explanation of this AI concept.
Learn termA simple explanation of this AI concept.
Learn termSee AI terms in action
Browse practical AI workflows that use the concepts in this glossary.
Last updated: