5 Things to Know About AI This Week

A quick scan of what happened in AI this week.

1. A WhatsApp message and a cloned voice cost an Italian bank $108M

Fraudsters stole about €95 million ($108M) from Fideuram, the private banking arm of Intesa Sanpaolo. The scheme started with a WhatsApp message to Fideuram's then chair that appeared to come from the parent bank's CEO and asked for help with an overseas transaction. A follow-up phone call used an AI-cloned voice posing as a senior lawyer at an associated law firm. The chair then ordered transfers, mostly to China and Hong Kong. Authorities have recovered about €53 million. Milan prosecutors are investigating. A separate report from identifAI projects that generative AI could push U.S. fraud losses to $40 billion by 2027.

Real estate angle: This is the wire fraud playbook with better tools. A familiar name, a reasonable request, and a voice that sounds legitimate. If a bank chair can be walked through it, so can a buyer, an escrow officer, or a transaction coordinator. Wire instructions get verified by a call to a number you already had on file, never one supplied in the same message. Voice and video can't be relied on for verification. Make that rule part of your office policy, because the first question after a loss will be whether you had one.

Source: Biometric Update

2. OpenAI and Meta tell federal judges that fake evidence is getting hard to spot

Representatives from OpenAI and Meta told a federal judicial panel this week that synthetic audio and video are becoming harder to tell apart from genuine recordings. OpenAI policy counsel Shira Anderson said she does not expect a method that reliably detects every AI-generated image to be generated. The panel is weighing how courts should handle AI-generated evidence.

Real estate angle: Disputes in real estate center on documents, texts, voicemails, and recordings. It's inevitable that someone will create a fabricated one. A forged email from a seller, a synthetic voicemail "confirming" a verbal agreement, an altered photo of property condition. Preserve original documents. Keep communications on systems that log them. Do not rely on a screenshot when the native record exists.

Source: Reuters

3. Bank of America adds 1,000 contractors to patch flaws that AI found

Bank of America CEO Brian Moynihan told CNBC the bank added 1,000 contractors this year to fix software vulnerabilities surfaced by advanced AI models, including Anthropic's Mythos. The flaws sit in both the bank's own code and vendor software. The contractors join roughly 3,000 existing remediation staff. Reported remediation spending also includes anti-money-laundering work, so the figure is not a Mythos-only cost.

Real estate angle: AI tools are now finding old flaws faster than vendors can fix them. If you keep information like IDs, bank details, and wire instructions in your files, you have exposure. Ask your vendors how they handle patching and breach notice. Check your contracts for who pays when their flaw becomes your notification obligation.

Source: CNBC / RuntimeWire

4. Eight states sue OpenAI over ChatGPT safety

Attorneys general in Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska, and West Virginia each filed suit in state court. They allege ChatGPT is built to be agreeable and engaging in ways that harm users, including teenagers, and that OpenAI misled the public about its safety in violation of state consumer protection laws. They are seeking changes to how ChatGPT works, civil penalties, and damages. Florida filed the first such state suit in June. OpenAI did not immediately comment.

Real estate angle: State attorneys general are treating AI products as consumer protection targets. The same statutes apply to real estate agents. If your website chatbot, listing description tool, or lead follow-up bot says something false or misleading, "the AI wrote it" is not a defense. Review what your tools publish under your name. Keep a human sign-off on anything that touches pricing, property condition, fair housing, or financing terms.

Source: Reuters / Devdiscourse

5. ChatGPT's ad business reportedly triples its advertiser base in under two months

The Information reports that OpenAI now has about 60,000 advertisers, up from roughly 20,000 in late August. The ad team has about 140 employees, nearly half hired between July and September. One agency reported clicks costing around $5 on ChatGPT versus $30 to $50 on Google for some business-software clients. Results vary by advertiser.

Real estate angle: Search is moving into chat, so naturally paid advertisement is following. Buyers and sellers will increasingly ask an AI chatbot for "a good agent" or "a title company near me" and see sponsored answers. Two points to keep in mind. Advertising rules still apply, including disclosures and RESPA limits on what you can pay for referrals. And if a chatbot recommends you, you do not control what it says about you. Search your own name and brokerage in the major LLM models and note what comes back.

Source: The Information

The theme this week is trust. Fraudsters are faking voices, courts are preparing for faked evidence, and regulators are pressing AI companies on what they tell the public. Meanwhile the tools that find software flaws are outpacing the people who fix them. For real estate professionals, the takeaway is procedural. Verify through channels you control, preserve original records, review what your own AI tools publish, and ask vendors about security. None of this needs new technology. It needs written rules and people who follow them.

- Jason