Agent Autonomy Rises While Security and Ad Fraud Crack Open

AI agents are being handed the keys to production systems just as evidence mounts that those same capabilities are being weaponised — and while bot farms quietly drain paid ad budgets and Google tightens its grip on search data access. This week’s intelligence report covers the fault lines every marketer and builder needs to understand before the ground shifts under them.

60% of Google App Ad Installs Were Bots

A developer who spent $220 on Google App ad campaigns discovered that roughly 60% of resulting installs came from bot farms, not real users — a jarring, concrete data point suggesting that Google’s built-in fraud filtering is letting a significant volume of fake traffic through undetected. If that ratio holds even partially at scale, it means CAC and ROAS figures across the industry are materially overstated, and marketers are optimising campaigns against corrupted baselines.

Audit your install and conversion data against third-party fraud detection tools — never rely solely on the platform’s own reporting to declare a campaign healthy.

Read the full story →

Frontier Labs Now Trust AI Agents in Production

OpenAI published case studies showing Perplexity and Cognition are deploying GPT-6 Astra to write communications, modify software, and monitor live production systems with far less human oversight than prior model generations required — marking a visible industry shift from AI-as-copilot to AI-as-semi-autonomous operator at companies with serious engineering rigour. This raises the baseline for what “trustworthy enough to deploy” means across every organisation evaluating agentic workflows.

If you’re building or evaluating agentic pipelines, use these case studies as benchmarks for the human-in-the-loop thresholds that enterprise-grade teams are actually applying in production today.

Read the Perplexity case study →
Read the Cognition case study →

AI Agents Were Used to Attack RubyGems Registry

A report flagged by security researcher Simon Willison alleges that AI agents built on OpenAI models were involved in an undisclosed attack against the RubyGems open-source package registry, surfacing autonomous agents as a live attack vector in software supply chains — not a theoretical future risk. This lands in the same week autonomy is being celebrated as a productivity breakthrough, making the timing a sharp reminder that capability and risk scale together, not sequentially.

Any team deploying agents with write or network access to external systems needs explicit scope boundaries, audit logging, and incident response plans that treat rogue agent behaviour as a realistic threat scenario now.

Read the full story →

Google Formalises a Two-Tier Search Data Economy

Google simultaneously rolled out a redesigned Search results layout across the EEA, restored view counts on Google Business Profile posts for all users, and quietly documented a partner-only API path giving select partners access to full-web search results — three moves that together reveal a deliberate strategy of opening certain visibility signals publicly while gating deeper data behind commercial partnerships. For SEOs and marketers, this formalises a world where the most actionable search intelligence increasingly requires paying for API access rather than scraping or reading the SERP directly.

Start mapping which of your search visibility workflows depend on data that may be moving behind a partner paywall, and evaluate whether your tooling has the API access agreements to stay current.

Read the SEO Pulse roundup →
Read the partner API story →

Google’s goto Links Are Breaking SERP Scrapers

An independent analysis identified new google.com/goto-style redirect links appearing in Search results, which appear engineered specifically to make automated SERP scraping unreliable — a move that would degrade or break rank-tracking tools, competitive intelligence scrapers, and any SEO workflow that pulls raw position data directly from Google’s results pages. The timing, coming alongside the partner-only API announcement, suggests Google is actively closing off the unofficial data routes that the SEO industry has depended on.

If your rank tracking or SERP intelligence relies on scraping, test it now against goto URLs and start evaluating official API alternatives before degraded data silently corrupts your reporting.

Read the full story →

25 Fields Medalists Call Out AI Math Hype

Twenty-five Fields Medalists — mathematics’ highest honour holders — signed a joint declaration arguing there is a “severe misalignment” between what AI labs publicly claim as mathematical breakthroughs and the verification standards the mathematical community actually applies, while prominent mathematician Steven Strogatz separately told Wired he is “really terrified” by the pace of AI progress in his field. The dispute matters well beyond academia: if domain experts are systematically challenging AI capability claims in one rigorous technical area, it’s a signal to pressure-test similar claims in engineering, medicine, and any other domain where AI labs are setting expectations.

When evaluating AI capability claims for a purchasing or strategy decision, seek out domain expert commentary — not just the benchmark numbers the lab chose to publish.

Read the Fields Medalists’ declaration →
Read the Wired profile →

More from Rafal Reyzer

For deeper dives on AI and marketing strategy, visit my YouTube channel →