Anthropic and OpenAI ask Australia to ease its ban on AI training with local content

Anthropic and OpenAI urged Australian lawmakers to revisit rules preventing AI companies from training models on the country’s creative works. Australia said in October 2025 that it would not adopt a text-and-data-mining exception, potentially increasing licensing requirements and legal constraints for AI model development in the market.

Analysis

Australia is a small direct revenue pool for frontier-model training, but its policy direction matters disproportionately as a common-law precedent for other English-language jurisdictions. A durable no-exception framework raises the marginal cost of model development through licensing, audit, and provenance requirements; that favors scaled incumbents with proprietary, rights-cleared data over model providers reliant on broad web scraping. For TRI, the relevant upside is not a one-time content license but stronger pricing power for Westlaw and Practical Law as customers place a premium on indemnifiable, traceable AI outputs.

The nearer-term risk is that a negotiated licensing regime becomes a de facto low-cost compulsory-access system, legitimizing training while limiting publisher economics. Over 1-3 months, this is primarily a sentiment and policy-optionality signal rather than an earnings catalyst; monitor whether submissions shift toward collective licensing, opt-out rules, or statutory remuneration. Over 6-18 months, a fragmented global regime could increase compliance costs for OpenAI and Anthropic, but also entrench their largest-capital advantage versus smaller model vendors—potentially reducing the value of policy protection for data owners if customer demand consolidates around a few platforms.

Consensus may overstate the benefit to all "content" equities. Generic publishers have weak bargaining power because their material is substitutable; TRI's relative advantage rests on workflow integration, editorial structure, and legal-data exclusivity, not simply copyright ownership. The thesis is falsified if TRI reports AI-product adoption without incremental ARPU or retention improvement, or if Australian legislation adopts a broad exception without meaningful recordkeeping, remuneration, or output-liability provisions.

Key Decisions for Investors

• Maintain a modest long TRI versus a short basket of ad-dependent/generic content publishers over the next 6-12 months; TRI should monetize regulated-data scarcity through legal workflow pricing, while generic licensors face weak negotiating leverage. Reassess if TRI's next two earnings reports fail to show AI-led net-sales or margin contribution.
• Do not add outright TRI solely on this policy headline. Set an alert for parliamentary draft language establishing mandatory provenance, opt-in licensing, or meaningful remuneration; those provisions would be a more credible catalyst for a 3-6 month multiple rerating.
• For AI-platform exposure, favor a barbell of scaled infrastructure beneficiaries over smaller foundation-model challengers if global licensing rules proliferate: higher compliance fixed costs are likely to widen the capital moat. Avoid treating Australian policy alone as sufficient evidence for a sector-level trade.
• Use any sharp TRI rally tied to a single licensing announcement to trim rather than chase unless management quantifies contract value, recurring revenue treatment, and margin economics; one-off training fees have materially lower quality than recurring workflow monetization.