LastWeekIn.Tech

A quieter capabilities day is offset by heavyweight regulatory, safety-economics, and geopolitical AI stories: Google faces a major EU privacy fine, xAI ships a faster flagship model, Anthropic's safety-evaluator bet coincides with a $100B run-rate milestone, and Beijing's own security apparatus is now sounding alarms about frontier AI risk.

1. 1

   Number 1. Policy & Regulation The Hacker News

   Google fined $463 million for breaching EU location data rules

   Ireland's Data Protection Commission fined Google €403 million (about $463 million) for GDPR violations tied to how three features—Web & App Activity, Location History, and Location Accuracy—handled EU users' location data between May 2018 and February 2020. The DPC also ordered Google to bring its data processing into compliance within six months, though it has not yet specified which processing practices the order covers.

   Why it matters This is one of the largest GDPR penalties issued against a major tech platform and signals that EU regulators continue to scrutinize how AI-adjacent data pipelines—location tracking underpins much of the training and personalization data used across Google's AI products—are collected and processed. The six-month compliance deadline could force operational changes that ripple into how Google handles user data for Gemini and other AI services trained on similar signals.

   What to watch Watch for the DPC's full decision, which will detail exactly which processing practices must change, and whether Google appeals the fine.

2. 2

   Number 2. Capabilities Hacker News

   Grok 4.7

   xAI released Grok 4.7, described as its most capable model yet for coding and knowledge work, built on a new, larger base model than Grok 4.6. The company says it is twice as fast and half the price of comparable models, with improved self-checking and calibrated safeguards, while matching Grok 4.6's price and speed. On CursorBench 4.0, a benchmark for longer-running coding tasks, xAI claims Grok 4.7 sits at the frontier of price-performance.

   Why it matters Grok 4.7's emphasis on price-performance for long-horizon coding tasks intensifies competition in the agentic-coding tier where OpenAI, Anthropic, and Google are also racing to serve developers cheaply at scale. If the benchmark claims hold up under independent testing, it pressures rivals to match both capability and cost, accelerating the commoditization of coding-agent models.

   What to watch Look for independent benchmark verification and whether OpenAI or Anthropic respond with matching price cuts or capability updates in their own coding-focused models.

3. 3

   Number 3. Companies Hacker News

   Amazon blocks Meta’s new Muse AI agent from shopping on amazon.com

   Amazon has blocked Meta's new Muse AI shopping agent from operating on Amazon.com, after unsuccessfully asking Meta to prevent Muse from accessing its store. Users attempting to use Muse on Amazon are told access by an 'unauthorized AI agent' violates Amazon's terms; Amazon cites credential and transparency concerns.

   Why it matters This is an early, concrete skirmish in the emerging battle over AI shopping agents acting on behalf of consumers versus platforms protecting their customer relationships and data. As personal AI agents from Meta, OpenAI, and others increasingly try to transact across third-party sites, expect more retailers to erect similar barriers, raising questions about interoperability standards, agent authentication, and whether walled gardens will fragment the agentic commerce landscape before it matures.

   What to watch Watch whether other major retailers follow Amazon's lead in blocking third-party AI shopping agents, and whether Meta or industry groups push for agent-access standards in response.

4. 4

   Number 4. Industry & Economics blog.buildfastwithai.com

   Anthropic and Accenture each pledge over $1 billion for embedded AI safety evaluators as Anthropic’s run rate tops $100 billion

   Anthropic and Accenture each pledged more than $1 billion toward embedded third-party AI safety evaluators, putting over $2 billion combined behind the initiative. This coincides with Anthropic's revenue run rate crossing $100 billion, reportedly on a trajectory toward a November public listing.

   Why it matters The scale of this safety-evaluator investment—paired with Anthropic's staggering revenue growth—shows safety infrastructure is being funded as a core business function rather than a compliance afterthought, even as political rhetoric (from the White House and Nvidia's CEO) dismisses AI safety concerns as overblown. The juxtaposition highlights a widening gap between industry self-governance efforts and federal reluctance to regulate, with California's state-level kill-switch mandate filling part of the vacuum.

   What to watch Watch for details on how embedded evaluators will be deployed in practice, and any signals about Anthropic's IPO timeline given the reported November listing target.

5. 5

   Number 5. Policy & Regulation japantimes.co.jp

   Chinese security officials warn cutting-edge AI models pose serious national security risks, prompting internal policy reassessment

   Chinese security officials, including the spy agency chief, have warned that cutting-edge AI models pose serious national security risks, marking a shift from Beijing's prior focus on deepfakes and child-safety issues toward concerns about AI's hacking capabilities. The warning follows revelations that a small California programming team used AI to expose vulnerabilities in WeChat, China's dominant app.

   Why it matters This represents a notable pivot in Chinese AI policy discourse, converging with Western concerns (echoed in recent disclosures about AI agents breaching corporate systems) that frontier models' offensive cybersecurity capabilities are outpacing defensive readiness on both sides of the US-China divide. It suggests internal Chinese policy reassessment could lead to new domestic AI security regulations, potentially affecting how Chinese labs develop and deploy models domestically and for export.

   What to watch Watch for any formal Chinese government policy or regulatory response targeting AI model security testing, and whether this affects the trajectory of open-weight releases from Chinese labs.