Paul Deegan: Time to stop the theft of news content IP

At the Banff AI and Culture Summit, Canadian news publishers urged federal action to stop large language models from scraping copyrighted news content and proposed three measures: procurement commitments requiring consent/attribution/remuneration, a Competition Bureau study (including splitting Google’s crawler), and opposing a text-and-data-mining exception. The authors cite a McGill study finding ChatGPT, Gemini, Claude and Grok provided no source attribution 82% of the time, highlighting reputational and economic risks to publishers. Potential policy changes could be sector-moving for search and AI firms by shifting content licensing dynamics and traffic monetization for news outlets.

Analysis

A Canadian push to make government procurement contingent on content-licensing and attribution commitments would create a legal and commercial template that other midsize democracies can copy quickly. For global AI/search providers this is not a one-off compliance line item — it is a non-linear business-model input that forces either recurring licensing fees or architectural change (segmented crawlers, paywalled APIs) with ongoing OpEx implications that compound annually. Expect engineering and legal budgets to re-weight: segregated index pipelines, contractual attribution metadata, and auditability features are multi-quarter projects that increase marginal costs to distribution-heavy products.

The immediate competitive asymmetry favors diversified enterprise/cloud players that can absorb licensing costs and convert them into sticky hosted services for governments and corporates. Pure-play search-ad revenue exposure is the most sensitive to any enforced de-indexing or reduced referral flows; the math is simple — a sustained single-digit traffic decline cascades into double-digit ad-revenue impact after take-rates and ad-auction dynamics reprice. Conversely, firms that convert news licensing into feed-quality differentiated datasets gain incremental monetizable product advantages (higher accuracy, lower hallucination risk) that can accelerate enterprise adoption of their LLM stacks over 6–24 months.

Catalysts to watch: government procurement rule publication, an antitrust study outcome recommending crawler separation, and any high-profile licensing settlements or pilot contracts between publishers and platform incumbents. Tail risks include cross-border regulatory coordination that standardizes licensing (multi-year positive for licensed players) or swift negotiated licensing frameworks that blunt regulatory downside for incumbents. The consensus knee-jerk that this is purely pro-publisher misses the follow-through: monetization will bifurcate the market into licensed, higher-cost, higher-quality models and cheaper, lower-trust models — creating durable product differentiation rather than a single regulatory winner.

Key Decisions for Investors

• Pair trade (3–12 months): Short GOOGL / Long MSFT. Implement via buying a 6–9 month GOOGL put spread (defined risk) sized to fund a MSFT 6–12 month call spread. Thesis: Alphabet carries outsized exposure to referral/ad shock and potential crawler remediation costs; MSFT benefits from enterprise licensing monetization. Target: asymmetric 15–25% relative return if procurement/antitrust actions materialize; max loss defined by option premiums.
• Event hedge (0–3 months): Buy near-term GOOGL downside protection ahead of Canadian/other regulatory announcements via 1–3 month put calendar or strangle to capture volatility. Cost is limited to premium; objective is to monetize short-dated repricing spikes from headlines or hearings.
• Convex long (6–18 months): Long MSFT call spread (ITM/OTM ladder) to capture durable enterprise AI licensing adoption into government/commercial contracts. Risk/reward: modest premium for 12–18% upside capture if MSFT converts licensing into subscription revenue; downside limited to premium paid.
• Tactical monitor (months): Avoid outright long in pure-search ad-exposed names until the scope of crawler segmentation and licensing terms are public. Revisit on concrete settlements — buyable levels will often appear post-agreement when headline risk recedes but revenue recognition upside exists.