Expert Witness Journal Issue 66 April 2026 - Flipbook - Page 80
example of a radiologist using (or failing to use) AI
tools to review medical imaging scans.
There may well be circumstances in which the use
of an AI tool gives rise to harm, but there is no
negligence or other direct liability on the part of
the developers of that tool. In those circumstances,
the fact that, under current English law, AI does not
clearly fall within the concepts of a legal person, it
may make it di昀케cult to ascribe liability to any party.
If an AI tool is not itself a legal person then (under
current English law principles) it cannot itself be
liable for harms caused to a user by its output.
There may also be questions as to how far “up the
chain” liability might travel where an AI model
produced by one party is incorporated into tools
developed by another party and in turn used by a
third party. As the Legal Statement notes, this will
be a highly fact speci昀椀c question, but questions of
liability in this regard in AI related claims are no
di昀昀erent to those arising from any other tool or
product.
One route for attributing liability to a legal person
might be by vicarious liability. However, a third party
(such as the provider or developer of AI tools) can
only be vicariously liable for the actions of another
legal person. For example, whereas a company may
be vicariously liable for the actions of one of its
employees, if an AI tool is not a legal person itself
then the operator of an AI cannot be vicariously
liable for its actions. So, for example, a company
may be vicariously liable for its employee’s negligent
use of AI, but cannot be vicariously liable for harms
caused by erroneous AI output provided directly to
a third party (e.g. a consumer).
However, negligence claims arising from the use
of AI will give rise to novel questions as to whether
the autonomous actions of the AI tool amount to a
break in the chain of causation. If this is the case,
the party that created AI model or tool may not be
liable for harms caused by its output. It may also give
rise to questions as to whether the behaviour of the
AI tool could and/or should have been anticipated
by the parties involved in developing it. Again, these
will be highly fact speci昀椀c issues, and are likely to
require expert evidence to resolve.
Given that AI tools can act with a degree of
autonomy, there is a potential gap in the law if there
are circumstances in which neither the AI itself nor
its operator can be held liable for harms arising
from the AI’s actions. As set out above, in many cases
there will be contractual allocations of liability, but
there may well be circumstances where this is not
the case.
False statements by AI
Another issue which the Legal Statement addresses
is who might be liable for harm caused by false
statements made by an AI.
The Legal Statement takes the view that such claims
will generally depend on whether the particular
statement made by the AI can be said to have been
made “by or on behalf of” a particular legal person.
As the UKJT takes the view that an AI cannot itself
be treated as a legal person, claimants would need to
show that the AI was making statements “on behalf”
of another legal person. The same principle would
apply to other claims based on statements (such as
claims in defamation or deceit).
Causation
Causation is another aspect of English law that may
give rise to complex issues when dealing with AI.
The ‘black box’ nature of AI decision making (i.e.
the fact that it is often impossible to discern the way
in which an AI reaches any particular output) may
make it di昀케cult to demonstrate who is responsible
for an output that has led to harm. This could
manifest in a lack of evidence (i.e. records of the
process undertaken by the AI) and/or opacity as
to how the AI operates (i.e. the actual reasoning
applied by the AI).
Again, this will be a highly fact-speci昀椀c consideration.
The authors of the Legal Statement acknowledge
that in many cases this is not something that a
claimant would be able to show. However, they
suggest that “given that AI is generally used as a
tool, the core negligence (if any) of the defendant
legal person would typically be the careless acts
that permitted the output, like the human decisions
behind the AI tool’s design, testing and deployment”.
In other words, the UKJT’s view is that in most cases
harms caused by false statements by AI will give
rise to negligence claims against the developers of
the relevant AI model or tool, rather than claims in
misstatement, defamation or deceit.
The Legal Statement takes an optimistic view
on this issue, arguing that these di昀케culties “are
neither more severe nor di昀昀erent in kind to the sorts
of issues that arise in other domains and that the
English common law is well able to accommodate”.
This may be the case. However, until such time as
the courts grapple with these issues, it remains to be
seen how much of an obstacle to claims they prove
to be. Unless and until there is judicial consideration
of these issues, there will inevitably be a degree of
uncertainty.
Vicarious liability
Another key consideration is whether legal persons
that develop AI models or tools which incorporate
AI can be vicariously liable for any harms caused by
that AI to its users.
EXPERT WITNESS JOURNAL
78
APRIL 2026
