AI News Digest, July 13: Europe’s AI Rules Get Real on August 2

Three weeks from now, EU AI Act enforcement stops being a line on a calendar. On August 2, the European Commission can start fining general-purpose AI providers. The Article 50 transparency rules also switch on across all 27 member states. So this matters to you if your team screens candidates, runs a customer chatbot, or ships AI content into Europe. It is the week to check where you stand. Below: the Act’s real demands, a Chinese lab building its own chip, and a quiet shift in the layoff story.

EU AI Act Enforcement Lands August 2, and Hiring Tools Sit in Scope

What changes on August 2

Two things switch on at once. First, the Commission gains real power over general-purpose AI models. It can open investigations, order corrective measures, and impose fines (Source: Tech Times). Second, the Article 50 transparency rules apply. That means disclosing when a person is talking to an AI system. It also means marking AI-generated content so machines can detect it (Source: EU AI Act text).

Here is the part people miss. Foundation-model providers have carried these obligations since August 2025, but the Commission could not fine anyone during that first year. From August 2, it can, and reportedly for conduct dating back to 2025.

Why AI Act compliance now reaches HR

Article 50 grabs the headlines, but the sharper edge for HR sits elsewhere in the Act. AI used to screen, rank, or evaluate job candidates counts as high-risk. So if you run AI hiring tools on EU-based applicants, the bar is already high. You need documentation, human oversight, and bias controls that many teams have not built yet.

And this is not a big-company problem. A 200-person startup with a few engineers in Berlin is exposed the same way a Fortune 500 is. Penalties for transparency breaches reach €15 million or 3% of global turnover, whichever is higher.

Getting ahead of EU AI Act enforcement

Start by mapping every AI touchpoint that reaches an EU user or candidate. Generative systems already on the market get until December 2 to meet the machine-readable marking rule. So you have a short runway, not zero. Meanwhile, if your AI applicant tracking stack scores candidates, ask the vendor for their conformity paperwork today rather than in October.

The Rest of Today’s Stories

A Chinese AI Lab Now Wants Its Own Chip

DeepSeek is developing its own inference chip to cut its reliance on Nvidia and Huawei, Reuters reported (Source: Engadget). The chip targets inference, the stage where a trained model answers users, rather than training. The lab has reportedly courted foundry and memory partners for about a year and hired chip engineers quietly.

So what? This is the same move OpenAI made with its Broadcom-built “Jalapeno,” and one Anthropic is now weighing too. For founders hiring AI talent across Asia, it signals where the roles are heading. Hardware, not just models. If you are hiring AI engineers in India or the wider region, silicon skills are about to get expensive.

The Layoff Story Quietly Changed

A CBS News analysis argues the bigger 2026 story is not mass AI replacement. Instead, it is a quiet pullback in entry-level hiring as employers push roles toward higher-skilled work (Source: CBS News). That squares with LHH data. 87% of HR leaders have run or plan layoffs in the next 12 months. And 78% now call them routine rather than one-off events (Source: LHH). HR Executive reached a similar “it’s complicated” verdict on the AI-jobs link (Source: HR Executive).

So what? For a small HR team, the shift is from “cut costs once” to “reskill continuously.” The same LHH report found 77% of companies offer redeployment programs, yet only 19% of employees notice them. That gap is a communication problem, and closing the AI skills gap in HR starts with telling people the programs exist.

An Open-Weight Model Comes for the Coding Agents

Zhipu AI, known overseas as Z.ai, released ZCode. The harness turns its open-weight GLM-5.2 model into an autonomous coding agent. The target is Anthropic’s Claude Code (Source: South China Morning Post). GLM-5.2 ships under an MIT license with open weights and a one-million-token context window.

So what? Open-weight agents that rival proprietary tools change the build-versus-buy math for engineering teams. Cheaper coding agents mean a smaller company can automate more of its backlog without signing a big vendor contract first.

Quick Hits

• Anthropic is in early talks with Samsung to manufacture its first custom AI chip on a 2nm process. The move follows OpenAI’s Broadcom-made “Jalapeno” chip. It is still early, and Anthropic says it may not proceed (Source: TechCrunch).
• Indian AI startups crossed $1 billion across 157 deals in the first half of 2026. That is up sharply year over year. Sarvam’s $234 million round was the largest (Source: Inc42).
• “The Reasoning Trap,” an ICLR 2026 paper, finds that training models to reason harder can increase tool hallucination instead of reducing it. Worth a read before you trust an agent with live actions (Source: arXiv).

What This Means for Your Team

EU AI Act enforcement is really a documentation problem wearing a compliance costume. You need to show who oversees each AI decision and how, especially in hiring. So if your HR stack cannot produce that trail on demand, an AI-ready system that logs the human-in-the-loop is worth a look. Asanify’s HRMS is built API-first for exactly this kind of oversight.

FAQ: EU AI Act Enforcement and AI in Hiring

When does EU AI Act enforcement start?

It starts on August 2, 2026. From that date, the European Commission can investigate and fine general-purpose AI providers. The Article 50 transparency rules also apply across all EU member states. Foundation-model obligations have existed since August 2025, but fines only become possible now.

Are AI hiring tools covered by the EU AI Act?

Yes. AI systems used to screen, rank, or evaluate job candidates are classified as high-risk under the Act. Companies using them on EU-based applicants need documentation, human oversight, and bias testing, even if the company itself is based outside Europe.

What are the penalties under the AI Act transparency rules?

Breaches of the Article 50 transparency obligations can cost up to €15 million or 3% of global annual turnover, whichever is higher. Generative systems already on the market have until December 2, 2026 to meet the machine-readable marking requirement.

Not to be considered as tax, legal, financial or HR advice. Regulations change over time so please consult a lawyer, accountant  or Labour Law  expert for specific guidance.