UK Forces Google to Let Publishers Opt Out of AI Search—Who Wins?

The story of the day

The UK’s Competition and Markets Authority (CMA) just ruled that Google must allow publishers to opt out of having their content scraped for AI Overviews—without penalizing their search rankings. The decision, buried in a 400-page interim report on AI competition, is a tectonic shift: it turns the "public web" into a negotiable dataset.

For builders, this means three things:

1. Search fragmentation is now inevitable. Publishers with paywalls (The Times, FT, Bloomberg) will opt out en masse, forcing Google to either degrade AI answers or cut side deals. Expect "premium knowledge gaps" in Overviews by Q1 2027.
2. The data-licensing gold rush accelerates. Publishers now hold leverage—Google can’t delist them for refusing scrapes, but it can offer cash or traffic incentives. Reddit’s $60M/year API deal starts looking like a floor, not an outlier.
3. Regulatory arbitrage becomes a feature. The ruling applies only to the UK. Google will geofence compliance, creating a two-tier search experience: opt-outs for British users, full scrapes elsewhere. Builders targeting global audiences will need region-specific data strategies.

The CMA’s move isn’t about privacy—it’s about market power. By decoupling AI training from search visibility, they’ve handed publishers a bargaining chip. The next domino? The EU, where the Digital Markets Act already has Google’s legal team in overtime.

New tools & launches

Anthropic Stock as Down Payment — San Francisco realtors now accept Anthropic equity (pre-IPO) for home purchases, per Wired. Valuation pegged at $30B+ in private trades. [Wired]

MITRE ATT&CK for AI Threats — Anthropic and MITRE mapped 12 months of AI-enabled cyber attacks into a new framework. Focus: prompt injection as the new phishing. [Anthropic]

Code Purchase Program — Google quietly buys Play Store app code (via third-party brokers) to train AI, per 404 Media. Devs report offers of $5K–$20K for repos. [404 Media]

Models & research

Math AI’s Gödel Problem

A Science paper warns that current AI systems—even those excelling at formal proofs—hit fundamental limits when tackling problems requiring self-reference (à la Gödel’s incompleteness theorems). Test case: AI fails to verify its own proof-generating code without human scaffolding. The kicker? Mathematicians aren’t alarmed—they’re relieved. "It means we still have a job," quipped one Fields Medalist. [Science]

Natural Language’s Expiry Date

Elicit’s founder argues that chat interfaces will die within 18 months, replaced by:

• Structured queries (SQL-like prompts for precise outputs)
• Agentic workflows (AI as a silent coordinator, not a conversationalist)
• Ambient UI (voice/gesture for consumer apps, APIs for pros)

The claim hinges on one data point: 78% of enterprise LLM queries now use templated inputs, not freeform text. [Elicit Blog]

Industry pulse

"This is the first time regulators have treated data as labor."

— Dina Srinivasan, Yale Law/AdTech researcher, on the UK’s opt-out ruling. She predicts publishers will form cartels to demand revenue shares from AI search, not just licensing fees. "The NYT’s lawsuit was a opening bid. This is the nuclear option." [The Verge]

"We’re buying code like it’s 2014 and we’re Facebook acquiring Instagram."

— Anonymous Google Cloud exec, on the Play Store code-purchase program. The strategy mirrors Meta’s early moves: vacuum up niche datasets before regulators notice. Difference? This time, the targets are developers, not users. [404 Media]

Stat of the day

$0.00 — The amount OpenAI refunds for expired API credits.

Per Reddit threads, unused credits vanish at expiration with no recourse, even for enterprise accounts. The policy, buried in ToS Section 7.3, nets OpenAI an estimated $12M/quarter in "breakage revenue"—unclaimed credits from churned devs. [OpenAI API Terms]

Get the full analysis—subscribe to Gradient Pro.