IEC eTech
article here
Software and AI systems become less secure with age. The issue is gaining
increasing momentum with the IEC developing much needed standards.
As digital systems age, they become harder to secure, harder
to maintain and increasingly vulnerable to exploitation. In sectors where
outages or cyber attacks can threaten public safety, the question of how to
retire software safely is now an issue of national resilience. While AI systems
are much more recent than old software packages, the issue still must be dealt
with as it could become a huge problem sooner than later. (For more on the
decommissioning of AI Systems, read this interview in e-tech.)
IEC and ISO have prepared standards dealing with both.
The scale of the problem is huge
A study published in November 2025 by WPI
Strategy, commissioned by Cisco, highlights the scale of the problem
across critical national infrastructure (CNI). According to WPI Strategy: “In
2020, nearly half of business network infrastructure globally was estimated to
be obsolete or ageing, making it harder to patch, harder to secure, and easier
to exploit.”
The report compares end‑of‑life (EOL) exposure across the
US, UK, France, Germany and Japan. It found that the UK has the highest
relative exposure to EOL systems. Japan exhibited the lowest exposure,
reflecting sustained investment in lifecycle management. In the US, 80% of
federal IT spend goes to maintaining existing systems.
Healthcare sector at risk from using outdated systems
Healthcare is one of the most exposed sectors. A 2025 report from one local trust within the UK’s National
Health Service outlines the risks: “The network infrastructure… has become
obsolete, lacking vendor support. This increases the risk of failure.” The
report confirms that legacy issues affect electronic patient records (EPR) and
the electronic prescribing and medicines administration (EPMA) as well as
clinical decision support systems.
It warns that failure of certain EOL components could be
“catastrophic” by impacting the whole of the organization. It is not without
precedent. A 2024 cyber
attack demonstrated how outdated systems amplify ransomware attacks,
leading to cancelled surgeries and huge financial losses.
This case is emblematic of a broader trend. Institutions
like hospitals often rely on outdated network hardware, unsupported operating
systems and legacy applications that cannot easily be patched or upgraded.
Legacy problems for critical infrastructure
Legacy systems can put public infrastructure at risk. While
grid systems increasingly rely on new AI models for processes such as load
forecasting and predictive maintenance by ingesting vast amounts of telemetry
from substations, sensors and grid assets, they also rely on legacy software
systems. If these are not retired appropriately, they can play a part in
producing inaccurate predictions that cascade into operational instability.
They also can be easily hacked, especially if they are older systems that are
not kept up to date with security patches and updates.
The US Environmental Protection Agency (EPA) issued an
enforcement alert in May 2024 urging water system operators to conduct
cyber security risk assessments. Software systems do not simply expire.
Instead, they become misaligned with new systems and data inputs. They may
continue operating long after they become outdated or their performance
degrades.
In hospitals, this could mean a triage model that once
performed well begins making unsafe recommendations. In electricity networks, a
forecasting model may become unreliable as consumption patterns shift.
Keeping the machine running
Retiring software isn’t a matter of deleting files or
shutting down servers. It’s a question of preserving national capabilities
whose functionality must endure long after the hardware they are operating on
has been replaced.
Hardware ages quickly; software, at least conceptually, does
not. But the two are inseparable. And in that tension lies one of the most
pressing challenges for modern infrastructure. “Even though software designers
try to reduce hardware dependency through abstractions, and standards help with
that, there is always some dependency. Changing hardware often forces changes
in software,” explains Sundeep Oberoi, Chair of ISO/IEC JTC 1/SC 7, the joint subcommittee between ISO
and IEC responsible for software and systems engineering standards.
While hardware churns, software persists: servers are
replaced every few years, networking technologies leapfrog one another, mobile
phones are frequently discarded because they can’t support the latest operating
system. “This is where technology debt accumulates,” Oberoi says. “Legacy
systems, such as COBOL-based tax platforms for instance, continue to perform
essential functions, but the ecosystems around them move on. Re‑architecting
them becomes risky, expensive and unavoidable.”
The industry’s answer is evergreening: treating software
renewal as a continuous process rather than a crisis-driven overhaul. “We’re
nowhere near that ideal, but it’s the direction of travel,” Oberoi agrees. The
hope is that AI will assist in evergreening by identifying dependencies,
mapping impacts, and guiding change. “Conceptually, software has always had the
ability to examine itself, self-replicate and change. This is inherent in the
idea of a universal machine. AI sharpens those techniques and makes them much
more powerful,” Oberoi adds.
But there are limits he acknowledges. “We’re nowhere close
to the point where AI decides something needs to change - such as requiring a
different processor - and can build that processor on its own,” he says.
“Biological systems contain enough information to assemble themselves but AI
systems do not. AI may help us realize what needs to be done. But the ecosystem
still has to do it.”
Standards are essential for decommissioning both AI and
software systems
International standards supported by rigorous lifecycle
management, proactive investment and transparent governance are one of the ways
of ensuring that critical infrastructure remains resilient, secure and
trustworthy, as technology evolves.
ISO/IEC/IEEE
15288 brings structure and consistency to the way organizations
engineer and manage AI systems from concept through maintenance and evolution,
including disposal. Recently published ISO/IEC/IEEE 12207 is
to software what ISO/IEC/IEEE 15288 is to AI systems and deals with retiring
software rather than disposal. ISO/IEC 5338 describes
the lifecycle of AI systems based on machine learning and heuristic systems. It
is based on ISO/IEC/IEEE 15288 and ISO/IEC/IEEE 12207 with modifications and
additions of AI-specific processes. These internationally recognized guidelines
define clear decommissioning triggers which could be due to regulatory changes;
replacement by a validated successor; end of business need; risk of cyber
security exposure; and model drift.
Yet another standard, ISO/IEC
8183, defines explicit “data decommissioning” and “system decommissioning”
of an AI systems’ stages to manage data and model artefacts
responsibly. This is essential in hospitals, where patient data must
be retained for legal reasons but protected from exposure. ISO/IEC 42001 requires
organizations to: “plan and manage the decommissioning of AI systems in a
controlled manner… ensuring that data and model artefacts are disposed of
appropriately.”
These documents define how engineered systems evolve from
conception to retirement. But even they have limits. Oberoi explains, “SC 7
models systems only to the extent they can be represented as information. It
doesn’t model the physics of a hard disk, only the way data is stored on it. As
a result, the committee has not yet confronted the question of what to do with
obsolete equipment - only how to preserve the information it contains.”
Of course, software is no longer the domain of large corporations with deep
process expertise. It’s built everywhere into startups, micro-businesses and
small development shops. That’s why SC 7 developed ISO/IEC
29110-7-1:2026 for Very Small Entities (VSEs). “Smaller entities don’t
necessarily have the competence or resources that large organizations do,”
Oberoi explains. The VSE work takes existing SC 7 standards for testing
frameworks and lifecycle models and scales them down without diluting their
intent.
Together, these standards give critical infrastructure
operators a structured way to plan for end‑of‑life and manage AI drift, to
retire systems safely and reduce cyber security exposure. Those charged with
decommissioning software must leave a clear audit trail for how the system
operated including its known limitations, the reasons for retirement plus
evidence of safe disposal.