MyIPO public consultation · Copyright Act 1987 Case docket · 9 August 2026
The authorities
Read the status before the holding.
Every case the submission cites, with its court, what it decided, and how
firmly it can be read. On what a trained model retains, every decision is now under appeal, in
London as in Munich. The four points where the courts divide are set side by side, with where
the submission stands on each.
Statuses as at 9 August 2026 · primary texts opened 9 August · certainty carried in the type
Type grade tracks certaintyFinal decided, no appeal pendingOn appeal direction, not settled lawInterim pending, ongoing, not yet decided
01 The docket
The cases, court by court
Each entry states the holding as verified against trade sources on
7 August 2026, and the single place the submission puts it to work.
United KingdomHigh Court, EWHC November 2025
Getty Images v Stability AI
The court largely rejected Getty's copyright claims. The training claim
was withdrawn for want of evidence that any training took place in the United Kingdom,
and the outputs claim after the offending prompts were blocked. Getty did not allege
that the model stores copies of the works, and the secondary infringement claim failed
on that footing. Permission to appeal was granted on 16 December 2025, reopening what
counts as an infringing copy.
In the submissionTopic 2.1 Box 3, for territoriality only: London
declined the jurisdiction Munich took. See divisions one and three.
On appealPermission granted 16 Dec 2025, secondary infringement
GermanyLG München I 42 O 14139/24 11 November 2025
GEMA v OpenAI
Memorisation of lyrics within model parameters is reproduction, in the
model and on output. The court accepted that the text and data mining exception can
cover AI training in principle; memorised retention exceeded its purpose. Injunction,
information and damages granted.
In the submissionFormerly carried the memorisation premise in topic
2.1. The submission no longer rests on it. See division one.
On appealOLG München translation read in part, 9 Aug
GermanyLG München I 42 O 763/25 31 July 2026
GEMA v Suno
First European ruling on a generative AI music tool. The court restrained
reproduction for training carried out in the United States, applied United States law and
rejected fair use. The provider is liable, not the user, and EU AI Act compliance is no
copyright defence. Penalty up to EUR 250,000 per breach.
In the submissionTopic 2.1 Box 3, the liability allocation. Its
provider-not-user holding is untouched by Getty. See divisions three and four.
On appealAppeal signalled status unchecked
United StatesN.D. Cal., Alsup J 23 June 2025
Bartz v Anthropic
Training on lawfully acquired books is "exceedingly transformative" and
fair use. Downloading pirated copies to build a central library is not. The settlement
that followed, approved 20 July 2026, paid USD 1.5 billion, about USD 3,100 per work
across more than 480,000 works.
In the submissionThe acquisition and use split: the
lawful-acquisition limb of the hybrid at 2.1 Box 4. See division two.
FinalSettledSettlement creates no precedent
United StatesN.D. Cal., Chhabria J June 2025
Kadrey v Meta
Meta won partial summary judgment on fair use, but the reasoning cuts the
other way: harm to the market matters more than the purpose of the copying, and a
transformative use earns no automatic immunity. The judgment openly criticises the weight
Bartz gives transformativeness.
In the submissionShows the American position split inside one
district. See division two.
DecidedPartial summary judgment no appellate answer on record
United StatesD. Del. February 2025
Thomson Reuters v Ross
Training an AI on copyrighted legal material was not fair use. The use was
commercial, not transformative in substance, and created a market substitute that harmed
licensing opportunities.
In the submissionThe third answer in division two, and a pre-filing
check: the appeal may produce the first appellate ruling anywhere on AI training.
On appealThird Circuit argued June 2026
United StatesSupreme Court 25 March 2026, unanimous
Cox v Sony
Contributory copyright liability narrowed. Knowledge of infringement plus
a failure to take reasonable steps no longer suffices. Secondary liability now requires
inducement, or a service tailored to infringement; internet access, with its substantial
lawful uses, is not. The Fourth Circuit was reversed.
In the submissionTopic 2.4 Box 1, the restraint direction on
automated takedown. See division four.
FinalConcurrence by Sotomayor and Jackson JJ
United StatesS.D.N.Y., MDL in discovery
New York Times v OpenAI
Not yet decided. The claims survived motions to dismiss in 2025, and the
court has ordered production of millions of ChatGPT output logs.
In the submissionLawful training does not guarantee safe outputs, so
the training question and the output question stay separate. Supports record-keeping as
the operative control at 4.3.
OngoingDiscovery
IndiaDelhi High Court interim order
ANI v OpenAI
Prima facie, storing works to train an LLM falls within fair dealing
under section 52(1)(a), and outputs using retrieval do not infringe. Public interest ran
as the fourth factor of the analysis, against an injunction, and the interim injunction
was dismissed. A third model alongside American fair use and the European exception,
from a common-law neighbour.
In the submissionTopic 2.1 Box 1, the open-word reading shared
with section 13(2)(a) of the Malaysian Act, and Box 4, the third direction. Entered
9 August.
InterimInjunction dismissed read 9 Aug
European UnionCJEU ruling expected September 2026
Like Company
Expected to address the text and data mining exception. It lands after
this consultation closes and before any amendment is finalised.
In the submissionA timing point only: the European answer arrives
while MyIPO is still drafting.
PendingNot yet decided
02 Where the courts divide
Where the courts answer both ways
Each division states the competing holdings and where the submission
stands. None of the four is left to be discovered by the reader of the submission, or by
MyIPO's advisers.
One. Does a trained model contain copies?
Munich · on appeal
GEMA v OpenAI
Memorisation within model parameters is reproduction, in the model and on output.
London · on appeal
Getty v Stability
Getty did not allege that the model stores copies, and the claim failed on that
footing. The appeal granted in December 2025 reopens the point.
Where the submission stands.On neither answer, and the appeals prove the choice right: both sides of the question
are now before appellate courts. The copying act is the assembly of the corpus, which
happens whether or not any trace survives in the parameters. Topic 2.1 rests on lawful
acquisition plus remuneration, and the memorisation premise is retired.
Two. Is training transformative?
Alsup J · final
Bartz v Anthropic
On lawfully acquired books, "exceedingly transformative" and fair use.
Chhabria J · same district
Kadrey v Meta
Market harm outweighs the purpose of the copying. No automatic immunity.
Delaware · on appeal
Thomson Reuters v Ross
Not fair use. A market substitute rather than a transformation.
Where the submission stands.The split is the reason not to import the American model. Two judges of one district
disagree in writing, a third court rejects the doctrine outright, and the first appellate
answer is pending at the Third Circuit. Malaysia has no fair use doctrine to receive any of
it. 2.1 Box 4 proposes a statutory route instead.
Three. Which court reaches foreign training?
Munich · jurisdiction taken
GEMA v Suno
Took jurisdiction over training carried out in the United States, through a venue rule
for collecting societies, and applied United States law.
London · training out of reach
Getty v Stability
The training claim failed for want of evidence that any training took place in the
United Kingdom. What survives on appeal is the infringing-copy question, not the
territorial one.
Where the submission stands.This is the reach step at 2.1 Box 3: a liability rule avoidable by relocating
training answers who but not where. Rome II Article 8 leaves that door open, and nothing in
the EU AI Act closes it. The submission asks MyIPO to say something about reach rather than
inherit the gap.
Four. Who answers for infringement?
Munich · primary liability
GEMA v Suno
The provider answers for its own output, not the user who prompted it.
Washington · secondary liability
Cox v Sony
An intermediary answers only for inducement, or for a service tailored to
infringement.
Where the submission stands.Not a true conflict, and 2.4 Box 1 distinguishes them: Suno concerns primary
liability for a provider's own output, Cox narrows secondary liability for a neutral
service. Together they carry the same direction, precise allocation over broad
intermediary duties.
03 Verification
How firmly each entry can be read
The standing caution
The primary texts are open. On 9 August the GEMA v OpenAI judgment was
read in part in English translation, the ANI v OpenAI conclusion and fair dealing sections
were read from the judgment on file, the Getty account was verified against the judgment
text, and Cox was checked against the decision record. The statutory quotes from the
Copyright Act 1987 and CPTPP Articles 18.68 and 18.69 are confirmed against the Act and
the treaty.
Still summary-derived: Kadrey, Bartz, Thomson Reuters, the Australian material and the
Rome II analysis. Enough to shape an argument, not enough for a pinpoint citation.
Moving before 14 August
Thomson Reuters v Ross: argued 11 June 2026, no ruling as at 10 August 2026, the date this
document is frozen for filing. The Third
Circuit may still deliver the first appellate ruling on AI training before the
deadline
Getty v Stability: checked 9 August. Permission to appeal granted 16 December 2025
on secondary infringement; watch the Court of Appeal
GEMA v OpenAI: checked 9 August, on appeal at OLG München. An English translation
of the judgment is available and would close the read-in-full gap
GEMA v Suno: appeal signalled, status still unconfirmed
Like Company: watch only, the ruling lands after the consultation closes
Case docket · MyIPO public consultationStatuses as at 9 August 2026Contact Print v1.0 · certainty carried in the type grade