cd /news/ai-safety/finding-the-bug-got-cheap-shipping-t… · home › topics › ai-safety › article
[ARTICLE · art-172668] src=dev.to ↗ pub= topic=ai-safety verified=true sentiment=· neutral

Finding the Bug Got Cheap. Shipping the Fix Didn't.

Security firm Calif reported that one of its researchers used OpenAI's Codex to find a critical Java flaw, nicknamed Java4Shell (CVE-2026-114938), in about an hour and produce a working exploit two hours later for under $200 in tokens; Oracle has shipped a fix and CISA urged immediate patching of internet-facing systems. Infraxio founder Justin Pennington argues the same week's string of outages at Microsoft, JPMorgan Chase, United Airlines and an FAA ground stop shows remediation, not discovery, is now the bottleneck, and urges small teams to keep a written inventory of the JVMs and services they depend on and push updates through a single rollback-capable path.

recommendation

Keep a written inventory of every service and embedded JVM you run, and push updates through a single path that can be rolled back.

more in playbook: Running AI in production
by read3 min views3 publishedNov 19, 2026

An AI-assisted researcher found a critical Java flaw in about an hour. The same week, a string of big outages traced back to ordinary updates. If you run software for a small team, the work that matters now is on the patching side.

By Justin Pennington, founder of Infraxio, Ponte Vedra. Originally published on penningtonj.com. Two stories landed in the same week, and I think they belong together.

The first: Oracle shipped a fix for a critical Java flaw nicknamed Java4Shell (CVE-2026-114938). Security firm Calif said its researcher found it in about an hour using Codex, with a working exploit two hours later, for under $200 in tokens, as Quartz reported. It affects OpenJDK 17.0.6 and newer, including Amazon Corretto, Eclipse Temurin, Microsoft's build and Oracle Java SE. CISA told organizations to patch immediately, internet-facing systems first.

The second: NBC News counted at least five major U.S. systems going down in a row, from Microsoft to a nationwide FAA ground stop to JPMorgan Chase and United Airlines. The explanations so far are boring. Microsoft traced its nearly 24-hour disruption to a software bug introduced in July plus traffic and capacity pressure, per Fortune. Chase pointed to a software bug in its on-premise systems. A senior administration official told reporters, per CNBC, that the incidents came from "upgrades of the systems, systems being patched, hardware being applied" and other unrelated events.

Put those side by side and the shape is clear. Discovery got fast and cheap. Remediation is still a change to a live system, and changes are what broke things.

Most small and mid-size businesses don't run Java on purpose. They run a ticketing tool, an accounting package, a build server or an identity product that has a JVM inside it. CISA's own guidance notes that Java runtimes may not show up as separate entries in a software inventory.

So the first question after a headline like this isn't "how do we patch?" It's "does this apply to us at all?" If you can't answer that in an afternoon, that's the real finding.

A workable inventory for a small team doesn't need a platform purchase. It needs, per system:

A spreadsheet is fine to start. The point is that it exists before the next alert, not after.

Patching slower isn't the answer. Patching with a plan is. The checklist I'd hand any operations team:

None of this is new. What's new is the tempo. If AI keeps finding serious bugs this quickly, patch cycles that used to come a few times a year will come a few times a month, and every one of them is a production change.

At Infraxio we build operations software, including IFX Hub, for teams around Jacksonville, St. Johns County and the rest of Northeast Florida. The rule we hold ourselves to is simple: keep a written list of every service a system depends on, and push updates through a single path that can be rolled back. It isn't exciting work. It's the difference between a bad Tuesday and a bad week.

How does your team find out whether a new CVE touches something you run? A real SBOM, a spreadsheet, grepping images, or waiting for the vendor to tell you? I'd like to hear what actually works at small-team scale.

Spend an hour this week on the list of what you run. It's the cheapest protection you'll buy this year.

Justin Pennington is the founder of Infraxio, a product studio in Ponte Vedra, Florida. More at penningtonj.com and infraxio.com/justin-pennington.

Sources: NBC News (Nov. 17, 2026), Fortune (Nov. 17, 2026), CNBC (Nov. 18, 2026), Quartz (Nov. 18, 2026), CISA alert (Nov. 17, 2026), via the original post.

── more in #ai-safety 4 stories · sorted by recency
── more on @oracle 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/finding-the-bug-got-…] indexed:0 read:3min 2026-11-19 · —