Opinion: Friday’s ground stop was a turning point for legacy systems
A five-hour national ground stop, blamed on old systems meeting new equipment, follows a run of similar failures. Why we expect more over the next 18 months.
· 5 min read
By Track49
Yesterday afternoon, Friday November 13, the FAA halted most departing flights across the United States for about five hours. The AP called it the longest nationwide ground stop in 25 years. FlightAware counted more than 9,000 cancellations on Friday, and thousands more today. Transportation Secretary Sean Duffy put it plainly: “This is a legacy system problem.”
We think yesterday will be remembered as a turning point. It was not the most damaging failure of an old system, but it was one of the few that most people felt directly. Engineers and auditors have described this risk for years. On Friday it became a canceled or delayed flight for potentially more than a million travelers.
What we know so far
The cause is still under investigation, and officials have given more than one account. On Friday, Homeland Security Secretary Markwayne Mullin said new technology being added to the system “was overwhelming the older system”, and that there was no sign of a cyberattack. Today Secretary Duffy pointed to a remote FAA facility where crews were replacing copper lines with fiber. He said a router involved in that work began flooding other routers across the network, though whether the router or its installation was at fault is not yet known.
Whatever the final report says, the pattern is familiar. The failure happened during an upgrade, at the point where new equipment met old.
It was not the first this year
On September 21, a circuit failed at the Philadelphia approach control facility, which also handles Newark’s airspace. The FAA had already scheduled that circuit for replacement. That morning, a construction crew had cut the backup fiber. Flights stopped at Newark, JFK, LaGuardia, Philadelphia and Teterboro.
On October 21, PHP released fixes for four bugs, filed as two CVEs, that had sat in its engine for more than a decade. AI helped find them, and attackers were already trying to exploit them before the fix shipped.
In each case, a long-standing weakness surfaced when something around it changed.
Why we expect more, and larger, failures
Our view is that failures like these will become more frequent over the next 18 months. We see seven reasons.
- Old systems are finally being changed, often in a hurry. Boards, regulators and governments now accept that running old systems is a growing risk, and modernization is being funded in both government and business. That awareness is overdue, but the response is often rushed: upgrades scheduled to meet a deadline rather than a test plan, and replacements chosen because they feel decisive. Every upgrade touches a system that was stable partly because nobody touched it. Friday’s failure happened during exactly that kind of work. The answer is not to stop changing old systems, but to change them in small, reversible steps, as we described in Modernize, don’t rewrite.
- Finding flaws has become cheap. Large, old codebases were protected by how expensive they were to audit. AI has removed much of that protection, for attackers as well as defenders. The PHP flaws are a clear example.
- Load is rising. AI features and automated traffic put more demand on systems than their designers planned for, and we see it in our clients’ traffic. Old systems tend to fail at their limits, and those limits are now reached more often.
- Support keeps ending. Python 3.10 reached end of life on October 1, and .NET 8 and 9 on November 10. Each date moves more production systems into the unpatched column. At the same time, the people who understand the oldest systems are retiring.
- Agents now use the screens built for people. Personal AI agents now drive browser interfaces and applications directly: they log in, fill in forms and step through workflows on a user’s behalf. xAI launched Grok Bot as a paid beta in August and Meta launched Muse in September. Both run each user’s agent on its own cloud computer with a full browser, built to work with sites that have no API. In our clients’ logs, this kind of traffic has grown sharply since August. These agents work at machine speed, in parallel sessions, retrying on every error and taking paths no tester tried. Many older systems quietly depend on a human pace. They enforce the order of a workflow in the user interface rather than on the server, validate only in the browser, have no rate limits on the pages people use, and assume one submission per click. An agent breaks each of those assumptions, without being malicious, simply by being efficient.
- Everything shares the same few dependencies. More organizations rely on the same clouds, identity providers, security agents and SaaS vendors. One faulty upstream change can now stop thousands of unrelated systems at once. In July 2024 a single CrowdStrike content update crashed about 8.5 million Windows machines, grounding flights and delaying hospital care. And on Thursday, the day before the ground stop, Microsoft 365 failed worldwide. Outlook was unreachable for more than 12 hours in places and Teams was disrupted, with reports from 74 countries. Government systems in Singapore and Japan were among those affected. Microsoft said it saw no sign of an attack and was rolling back a network change, and officials said the two events were not related. Neither failure was in an old system, but old systems are rarely designed to route around a dependency that fails.
- AI-written patches land on code nobody understands. Under pressure to change old systems quickly, teams use AI assistants to extend codebases faster than anyone can review them. The changes compile and pass whatever tests exist, but the people approving them often don’t fully understand the system they are changing. We wrote about this kind of comprehension debt in September.
Governments are behind on this too. In June 2019 the US Government Accountability Office named the ten federal legacy systems most in need of modernization. By February 2025, three had been modernized. In September, Abigail Bradshaw, Director-General of the Australian Signals Directorate, said Australia is carrying “an enormous legacy technology debt”. Most private organizations are in a similar position, with less public scrutiny.
The risk in the reaction
After a failure this public, the pressure is to replace everything at once. Big-bang replacements reopen every decision the old system had already settled. They ship every new defect on the same day, and they leave no route back when something goes wrong. We expect some of the next failures to come from exactly those rushed replacements. We expect the rest to come from old systems that stay in place and are pushed past the load, security and uses they were designed for.
Sources
- Nextgov/FCW, Legacy tech issue led to FAA’s nationwide ground stop, November 13, 2026.
- AP via Federal News Network, Trump administration blames ‘overloaded’ systems for ground stop after FAA lets flights resume, November 13, 2026.
- TIME, Air Traffic Control Systems Are Outdated. What’s Being Done to Prevent Another Outage?, November 14, 2026.
- AeroCorner, A Failed Circuit and a Cut Backup Cable Stopped Flights at Five Northeast Airports, September 22, 2026.
- Wiz Research, PHP2Shell: AI uncovers critical RCE vulnerabilities affecting 70% of the Internet, October 21, 2026. See also our PHP2Shell analysis.
- US GAO, Information Technology: Agencies Need to Plan for Modernizing Critical Decades-Old Legacy Systems (GAO-25-107795), July 17, 2025.
- Python Software Foundation, Python 3.10.22 release; Microsoft, .NET support policy.
- Vellum, Official Grok Bot Breakdown (2026), August 20, 2026.
- KDnuggets, Meta Muse Explained: What It Is, How It Works, and What It Can Do, October 5, 2026.
- TechTarget, Microsoft: Faulty CrowdStrike update affected 8.5M devices, July 2024.
- Al Jazeera, Microsoft hit by global outages: What happened and why?, November 13, 2026.
- endoflife.ai, ASD Warns Legacy Technology Will Be First Hit in AI-Enabled Attacks, September 19, 2026, reporting Cyber Daily, September 16, 2026.
- legacy
- risk
- modernization
- security