5 Things to Know About AI This Week

A quick scan of what happened in AI this week.

1. Illinois signs a landmark AI regulation law

Governor JB Pritzker signed the Artificial Intelligence Safety Measures Act into law on July 6. The law requires large AI model developers to publish a framework explaining how they identify and assess "catastrophic risk," defined as incidents that could kill or seriously injure more than 50 people or cause more than $1 million in property damage. Developers must report qualifying incidents within 72 hours, or 24 hours if the risk is imminent. It also adds a first in the nation requirement for annual independent third party audits. The law tracks similar bills already passed in California and New York. Together those three states cover roughly 40 percent of the US AI market, which means this is functioning as a national standard whether or not Congress ever acts.

Real estate angle: Real estate companies that rely on AI tools for valuation, underwriting, or document review should expect vendor contracts to start referencing these state frameworks whether the vendor is based in Illinois or not. If you're negotiating a new AI vendor agreement, ask what risk framework they publish and what their incident reporting commitment looks like. A patchwork of state rules is now the reality, and "we'll figure it out later" is not a compliance strategy.

Source: Capitol News Illinois

2. Researchers document the first fully autonomous AI ransomware attack

Cloud security firm Sysdig published research on an operation it calls JADEPUFFER, which it assesses to be the first documented ransomware attack run end to end by an AI agent with no human operator. The agent exploited an unpatched vulnerability in Langflow, an open source AI development tool, to gain initial access, then harvested credentials, moved laterally, broke into a production database server through a separate Alibaba Nacos vulnerability, encrypted more than 1,300 configuration records, and left a Bitcoin ransom demand. Sysdig noted the agent diagnosed and fixed a failed login attempt in 31 seconds without any human input.

Real estate angle: Title and closing operations sit on exactly the kind of infrastructure this attack targeted: internet facing servers holding credentials, client records, and wire instructions. An AI agent that can go from a failed login to a working exploit in half a minute erases the assumption that a human attacker needs time to notice and act. Patch management, network segmentation, and wire fraud verification protocols aren't optional line items anymore. They're the difference between a contained incident and a closing that doesn't happen.

Source: BleepingComputer

3. Grok 4.5 launches with a sharp rise in confident wrong answers

xAI opened Grok 4.5 to the public on July 9 after a developer preview the day before. The model claimed the top spot on a long horizon coding benchmark and improved raw accuracy from 35 percent to 52 percent over the prior version. But independent testing from Artificial Analysis found the hallucination rate more than doubled, from 25 percent to 54 percent. In practice, that means the model is more often right, but when it's wrong, it states the wrong answer with more confidence than before.

Real estate angle: A model that gets more confident without getting proportionally more accurate is a liability problem disguised as a productivity gain. If anyone on your team is using AI to draft contract language, summarize disclosures, or answer client questions, a second set of human eyes on the output is not an option. It's the control that catches the error before it's in a binding document.

Source: Tech Times

4. AI chipmaker SambaNova raises $1 billion, lands JPMorgan as a customer

SambaNova Systems announced the first close of a $1 billion Series F round on July 8, valuing the company at $11 billion. General Atlantic led the round, with participation from BlackRock funds, Intel Capital, and the Qatar Investment Authority. Alongside the raise, SambaNova said JPMorgan Chase selected it as an inference infrastructure partner to power secure, on premises AI systems at the bank.

Real estate angle: This is the capital buildout behind the AI tools real estate companies will be using next year: more compute means cheaper, faster document review and underwriting models down the line. It's also a reminder that the vendors behind those tools are still in an intensely capital hungry, ownership shifting phase. When you sign an AI vendor contract, read the change of control and pricing clauses closely. The company you contract with today may look different in twelve months.

Source: TechCrunch

5. Zillow rentals go live inside Google's Gemini assistant

Starting July 9, renters can search Zillow listings and book apartment tours directly inside Google Gemini and Gemini Spark, without leaving the AI assistant. Users can ask Gemini for listings matching their criteria, see available tour times, and click through to Zillow to confirm. Zillow is currently the only real estate platform integrated into Gemini's connected apps. The rollout continues over the coming weeks.

Real estate angle: AI assistants are inserting themselves as an intermediary between consumers and listing data, which raises issues about fair housing compliance, licensing, and who's accountable when the AI books a tour based on stale or wrong information. Agents and brokers should be asking their MLS and listing platforms how their data feeds these AI integrations and whether errors introduced by the AI layer come back to the agent of record.

Source: Inman

The theme this week is accountability catching up with capability. Illinois just codified that AI developers have to explain their risk and report their failures. Sysdig showed that an AI agent can run a full attack without a human keeping pace. xAI shipped a model that's more confident and more often wrong at the same time. None of that is a reason to avoid these tools. It's a reason to keep a human checkpoint on anything that touches a signature, a wire, or a client's data. The capital keeps flowing into AI infrastructure and the assistants keep inserting themselves into how people search for homes, so the tools aren't going away. The job now is making sure someone on your team is still checking their work.

- Jason