July 30's AI news was a day when the arrows pointed in different directions across technology, regulation, and industry. While a classic technique for underpinning AI agent reliability is being reassessed, the US moved to block imports of Chinese robots in the name of protecting AI infrastructure, and the cloud market saw acquisition activity. Rather than a major rupture, the overwhelming impression is of multiple currents advancing at once.
Ontologies Are Back — "Logical Guardrails" for AI Agents
What caught the eye first was the move to reassess **ontologies** as a means of keeping AI agents in check. Frank Coyle, who teaches generative AI and LLMs at UC Berkeley, made the case at the recent AI Engineer World's Fair. While LLMs excel at probabilistic reasoning, running agents in earnest requires "logical guardrails" — and the guardrail he brought up was ontology.
An ontology describes the data structure of a given domain in terms of "classes, properties, and relationships." Coyle defined it plainly as "a way of representing data as a graph," noting that the concept itself dates back to Aristotle. The idea is to counter the tendency of LLM outputs to drift, reining them in with rules that machines can enforce.
Emil Eifrem, CEO of graph database company Neo4j, also cited three kinds of ontologies as ingredients for building a "smarter shared foundation" for running agents at scale: one for business, one for technical (metadata) purposes, and one for execution traces. Coyle calls this combination of neural networks and symbolic AI "neurosymbolic AI" and positions it as a means of keeping LLMs on track.
What's interesting is that the **technologies that once underpinned the Semantic Web** — Schema.org, FOAF, Dublin Core, and OWL — can be reused as-is. Because they are already present in LLM training data, the argument goes, invoking them via prompts is more practical than redefining them from scratch. The remaining challenges are maintenance and obsolescence — precisely why the Semantic Web never achieved broad adoption — though some argue that if agents themselves can update definitions as part of operations, the very nature of the problem changes. On the enterprise side, a survey was reported showing that more than half of companies attribute AI's wrong answers to missing business context, a result that bolsters the case for investment in semantic layers.
FCC Moves to Block Imports of Chinese Robots and Power Inverters
The US Federal Communications Commission (FCC), aiming to protect American AI infrastructure from foreign threats, has announced plans to block imports of new Chinese humanoid robots and robot dogs. However, the rule's definitions are broad and could sweep in cleaning robots like the Roomba, lawn-mowing robots, and delivery bots, raising concerns that the impact will extend further than anticipated.
Nscale Acquires Anyscale for Approximately $1.65 Billion
Nscale, a cloud provider for AI, announced that it is acquiring software startup Anyscale. The deal, valued at approximately $1.65 billion, is aimed at helping customers use AI compute more efficiently. Anyscale is known for the distributed computing framework "Ray"; the move can be read as a play to boost compute efficiency by combining hardware and software.
GCC and OpenJDK Publish Interim Generative AI Policies
Major open source projects laid out their AI stances simultaneously. The GCC steering committee announced an AI policy, and OpenJDK published an interim policy on generative AI. How projects should accept AI-generated code — including questions of copyright, quality, and attribution — is a topic of intense community interest, and GCC's announcement drew more than 100 points on tech communities, sparking discussion.
Amazon Confirms Cases of AI Sending Tech Project Spending "Off the Rails"
Amazon has reportedly confirmed internal cases where AI inflated the cost of technical projects beyond expectations. Details on specific amounts and case counts are limited, but it once again threw into relief the risk that, as enterprises deploy AI in the field, cost management fails to keep up on the flip side of productivity gains.
European Banks Chase AI Boom Profits While Keeping Risks in Check
Major European banks are signaling a willingness to capture the lending demand generated by the AI-related data center investment boom. Société Générale disclosed that its exposure to data center infrastructure amounts to approximately €7.7 billion (about $8.8 billion), while also moving to spread risk through hedges on project financing exceeding $5 billion. It's a picture of banks rushing to balance profit opportunities with exposure management.
---
**Note:** Meta disclosed in regulatory filings that future AI data center-related lease obligations not yet reflected on its balance sheet reached $279 billion at the end of the quarter, up 53% from the prior period. The weight of AI spending continues, with some observers noting that free cash flow for the quarter fell sharply.