AI Copyright Training

State of play.

• Anthropic is litigating fair use while simultaneously settling at scale. In California federal court, Anthropic has filed for a ruling that Claude's training on copyrighted materials constitutes transformative fair use -- analogizing AI ingestion to human learning -- while the $1.5 billion Bartz v. Anthropic class settlement, covering more than 100,000 authors and rights holders with 91% participation, moves toward a fairness hearing (→ Anthropic argues Claude's copyright use is transformative fair use in CA court).
• The UK has reversed course on a broad AI copyright exception. The UK government dropped its proposed opt-out regime and published a report signaling a policy pivot -- leaving the framework for AI training rights in the UK unsettled and under active reconsideration .
• PAE acquisition of orphaned AI-sector patents is a structural exposure vector. Holland & Knight maps a recurring pattern: failed AI startups leave foundational patents that migrate to assertion entities, which then target the companies that successfully commercialized the underlying technology .
• The federal AI legislative framework is shaping sector-specific compliance obligations, with the White House national AI legislative framework and the America AI Act creating downstream implications for how AI training and deployment are regulated, including in healthcare .
• For counsel advising AI developers, publishers, or rights holders, the practical baseline is that no settled fair use standard for large-scale training exists -- Bartz may establish a damages benchmark, but Anthropic's parallel fair use motion means the doctrinal question remains live even as settlements accumulate.

Where things stand.

• Bartz v. Anthropic is the leading settlement benchmark. The $1.5 billion class settlement covers claims from over 100,000 authors and rights holders; the claims deadline has passed and a fairness hearing is pending in San Francisco federal court (→ Anthropic argues Claude's copyright use is transformative fair use in CA court).
• Anthropic's transformative fair use argument is before a federal judge. The company's position -- that AI training is analogous to human learning and therefore transformative -- is the central doctrinal test for large-scale training use; no court has yet adopted or rejected this framing (→ Anthropic argues Claude's copyright use is transformative fair use in CA court).
• Publisher registration failures are surfacing as a litigation complication. Some publishers, including Macmillan, failed to properly register works before those works were ingested into training datasets, creating gaps in standing and damages calculations that complicate both plaintiff and defense strategy (→ Anthropic argues Claude's copyright use is transformative fair use in CA court).
• The UK has abandoned its opt-out AI copyright exception. The UK government's published report signals a policy reversal, leaving rights holders and AI developers operating without a clear statutory framework for training data in that jurisdiction .
• PAE risk is structural, not episodic, in AI-adjacent IP portfolios. As AI technologies mature from development to revenue generation, orphaned startup patents are migrating to assertion entities -- a pattern Holland & Knight identifies as intensifying as secondary patent markets have matured .
• The federal AI legislative framework introduces sector-specific training and deployment obligations. The White House national AI legislative framework and America AI Act create a compliance layer that intersects with copyright questions, particularly in regulated sectors like healthcare .

Latest developments.

• Anthropic's transformative fair use motion filed in California federal court, arguing Claude's training is analogous to human learning; Bartz v. Anthropic fairness hearing pending with 91% claimant participation in the $1.5 billion settlement (→ Anthropic argues Claude's copyright use is transformative fair use in CA court).
• UK government drops its proposed AI copyright opt-out exception and publishes a report signaling a policy pivot, leaving the UK training-data framework unresolved .
• Holland & Knight analysis maps PAE acquisition of orphaned AI startup patents as a structural litigation risk for companies commercializing AI technology .
• White House national AI legislative framework and America AI Act create sector-specific compliance obligations with downstream implications for AI training and deployment .

Active questions and open splits.

• Whether large-scale AI training qualifies as transformative fair use. Anthropic's pending motion is the most direct test of this question in federal court; the analogy to human learning has not been accepted or rejected, and no circuit has addressed it (→ Anthropic argues Claude's copyright use is transformative fair use in CA court).
• Whether Bartz settlement approval sets a damages benchmark. If the $1.5 billion settlement is approved, it will be the first significant reference point for per-work or aggregate damages in AI training copyright disputes -- shaping every subsequent negotiation and litigation (→ Anthropic argues Claude's copyright use is transformative fair use in CA court).
• Publisher registration failures and standing gaps. Works ingested without proper registration create standing and damages complications for plaintiffs; the extent to which courts will allow claims to proceed on unregistered works is unresolved (→ Anthropic argues Claude's copyright use is transformative fair use in CA court).
• UK training-data framework post-opt-out reversal. With the broad exception abandoned, the UK's path forward -- whether toward a licensing regime, a narrower exception, or continued uncertainty -- is genuinely open, with significant implications for cross-border AI development .
• PAE exposure from AI startup patent portfolios. The structural dynamic Holland & Knight identifies -- foundational patents migrating to assertion entities as AI companies mature -- raises unresolved questions about freedom-to-operate diligence and indemnification obligations in AI M&A and licensing .
• Federal preemption of state AI training obligations. The White House framework's posture toward state-level AI regulation creates uncertainty about whether state copyright-adjacent training requirements survive federal preemption analysis .

What to watch.

• The Bartz v. Anthropic fairness hearing -- whether Judge Martinez-Olguin approves the settlement and what damages framework the approval order articulates.
• The California federal court's ruling on Anthropic's transformative fair use motion -- the first major judicial test of the human-learning analogy for AI training.
• UK government follow-on action after the opt-out reversal -- whether a licensing regime, narrower exception, or continued gap emerges.
• Additional publisher or author class actions revealing registration failures or contractual disputes over AI training rights, which will complicate the litigation landscape even as Bartz moves toward resolution.
• PAE acquisition activity targeting orphaned AI startup patent portfolios as the sector matures -- watch for assertion campaigns against major AI developers.