Source: https://agentshelf.com/blog/5-things-ai-week-of-june-26-2026
Published: June 26, 2026
Author: Jason Hernandez
Category: Industry News

# 5 Things to Know About AI This Week

A quick scan of what happened in AI this week.

## 1. OpenAI built its own chip

OpenAI and Broadcom unveiled Jalapeño on June 24, OpenAI's first custom-designed AI accelerator. Engineering samples were hand-delivered to Sam Altman and Greg Brockman by Broadcom CEO Hock Tan at OpenAI's San Francisco headquarters. The chip was designed from scratch in nine months, with OpenAI's own AI models helping speed up the design cycle. Jalapeño targets LLM inference specifically, promising roughly 50% cost savings over standard NVIDIA GPUs. Initial deployment is planned for late 2026, with full ramp in the first half of 2028.

**Real estate angle:** This is about what it costs to run AI, which sooner or later shows up in the price of every platform your brokerage pays for. Cheaper inference means the tools built on OpenAI's models get cheaper to operate, and that pressure should eventually flow down to pricing. When your vendors renew, ask whether any of that savings is coming back to you.

*Source: [TechCrunch](https://techcrunch.com/2026/06/24/openai-unveils-its-first-custom-chip-built-by-broadcom/)*

## 2. Anthropic told the U.S. Senate that Alibaba stole Claude

In a letter to the Senate Banking Committee dated June 10 and first reported by Bloomberg on June 24, Anthropic accused Alibaba's Qwen lab of running 28.8 million unauthorized interactions with Claude between April 22 and June 5, 2026. The operation used roughly 25,000 fraudulent accounts and commercial proxy services to get around geographic restrictions. The goal was to extract Claude's most commercially valuable capabilities, including agentic reasoning and long-horizon task completion. It is the largest documented AI model distillation attack on record, exceeding the combined total of the three Chinese labs Anthropic reported in February.

**Real estate angle:** Fake accounts and proxies used to pull value out of a system at scale is the same pattern behind the wire fraud and identity theft schemes that target closings. Bad actors probe for gaps in access, often for months before anyone notices. There is also a vendor question here: if a tool you rely on was built on stolen data, that becomes your due-diligence problem. Ask your AI vendors where their models came from. Anthropic is asking Congress for export controls on AI model access and mandatory screening of high-volume API usage.

*Source: [Bloomberg](https://www.bloomberg.com/news/articles/2026-06-24/anthropic-accuses-alibaba-of-illicitly-accessing-its-ai-models) / [Tom's Hardware](https://www.tomshardware.com/tech-industry/artificial-intelligence/anthropic-claims-that-chinas-alibaba-illicitly-distilled-its-models-from-april-to-june-2026-says-effort-involved-25-000-fake-accounts-and-28-8-million-exchanges-on-claude)*

## 3. Deepfake identity fraud is on track for a 495 percent increase this year

A June report from identity verification firm Shufti Pro projects a 495% jump in deepfake identity fraud in 2026 over 2025. Document deepfakes are growing even faster, up an estimated 3,892% year over year. These are not phishing emails with bad grammar. Deepfake tools can now simulate live video notarization sessions well enough to fool human reviewers. One documented attempt in Maryland involved an AI-generated seller impersonation during a video notarization on a vacant land transaction worth roughly $100,000. Proof's detection software caught it. NAR has published a consumer-facing guide on spotting deepfake scams in real estate, and 47 states now have some form of deepfake legislation on the books.

**Real estate angle:** Video deep-fakes are the exact attack surface here. If you use a video call to verify identity, your video platform needs active deepfake detection, not just a good camera. Ask your platform and your E&O carrier about the Maryland case. Title companies and settlement agents carry real liability if a closing is later challenged because a deepfake beat the identity check, so put the detection requirement in the vendor contract.

*Source: [ASIS International](https://www.asisonline.org/security-management-magazine/latest-news/today-in-security/2026/june/deepfake-identity-fraud/) / [Inman](https://www.inman.com/2026/06/04/deed-fraud-real-estate-technology-crime/)*

## 4. Warner Bros. Discovery rebuilt its ad-tech stack around agentic AI

Warner Bros. Discovery announced on June 19 that it is replacing its advertising technology infrastructure with an agentic AI platform built on Amazon Web Services. The new stack automates media planning, audience forecasting, measurement, attribution, order management, pricing, and campaign stewardship using autonomous AI agents running on Amazon Bedrock AgentCore. Phase one is live now, covering direct response and commercial workflows. Unified media planning rolls out in Q3, with order management and pricing following in Q4.

**Real estate angle:** WBD is one of the largest ad buyers, and it is replacing human-run ad operations with AI agents. The marketing platforms your brokerage uses for social ads, email, and MLS syndication are heading the same way, so campaigns that used to need a coordinator setting them up and watching them will increasingly run on their own. That saves time. It also raises a question you should answer now: who is responsible when an autonomous campaign violates fair housing rules?

*Source: [Marketing Dive](https://www.marketingdive.com/news/warner-bros-discovery-revamps-ad-tech-stack-around-agentic-ai-aws/823252/) / [Sports Video Group](https://www.sportsvideo.org/2026/06/22/warner-bros-discovery-and-aws-announce-ai-powered-advertising-technology-platform/)*

## 5. The EU picked a consortium to build a sovereign AI model in all 24 official languages

On June 19, the European Commission selected the EUROPA consortium, led by Italian startup Domyn, as the winner of the Frontier AI Grand Challenge. The project will produce an open-source frontier AI model with more than 400 billion parameters, trained in all 24 official EU languages. The winner gets up to 2.5% of EuroHPC computing capacity for a year. The challenge was launched in February 2026 as Europe's answer to U.S. and Chinese AI dominance. Because the model will be open-source, any organization in the EU can use it without paying licensing fees to an American or Chinese provider.

**Real estate angle:** The EU AI Act's transparency rules go into effect in August 2026, and European real estate platforms are already building compliance around them. A sovereign open-source model gives them a foundation without data residency or contract questions about U.S. vendors. As with every story concerning how the EU is addressing AI regulation, this gives a hint of what could be coming our way. Disclosure requirements like these have a way of crossing the Atlantic.

*Source: [European Commission](https://digital-strategy.ec.europa.eu/en/news/commission-selects-europa-consortium-winner-frontier-ai-grand-challenge-project-build-european-open) / [AI Weekly](https://aiweekly.co/alerts/domyn-led-europa-consortium-wins-eu-frontier-ai-grand-challenge)*

---

The theme this week is accountability: who answers when AI is stolen, when it is used to commit fraud, or when an AI agent makes a decision that causes harm. OpenAI's chip is the infrastructure story, but Alibaba's alleged theft, deepfakes beating identity checks at closings, and autonomous agents running ad campaigns all point to the same pressure. The tools are ahead of the rules. Real estate sits where identity, money, and legal documents meet, so we are an early target whenever fraud changes. For real estate professionals, that means two things: check whether your video platform has real deepfake detection, and ask your vendors where their AI came from. Both will matter more over the next 12 months than the last 12.

_- Jason_