"Claude just gave me access to another user's legal documents"
Claude supposedly fed a Reddit user another user's lease, complete with real company names, a real property address, and what looked like sensitive deal terms.
But it turns out, according to many of the 2k+ comments, Claude did NOT share another user's legal documents. It committed a "high-fidelity fabrication." Real company names, real address, invented attorneys, invented terms. i.e., Claude made it up.
Either way, this brings up what is quickly becoming a key practical limitation of AI in the workplace: security.
---- The real issue? ----
We're running what we believe is the first comprehensive survey of AI adoption across the real estate industry, and early results confirm a clear pattern: AI power users exist, but they're concentrated at smaller firms or using tools outside the office.
Why? Because compliance won't let them in the front door.
The most popular tools aren't allowed at most firms.
---- Why the fuss? ----
Information security risk scales with your industry.
Nobody's career ends if a dry cleaner loses their phone number. But if you're a broker who leaks a client's sensitive deal terms, you're done. If you're an attorney who fabricates a document and someone relies on it, you're done faster.
Real estate and financial brands don't survive on average. They're fine until they're not. Arthur Andersen had 85,000 employees and 89 years of history. One engagement ended all of it.
---- The compliance paradox ----
Compliance attorneys often disagree among themselves about what's actually risky. E.g., Ares has comments locked down on LinkedIn, while Blackstone doesn't. Same industry, same regulatory environment, very different interpretations.
Are large firms with conservative compliance teams protecting themselves or falling behind while smaller competitors move faster?
Are smaller firms without those guardrails at an advantage, at least until they do something stupid?
COMMENTS