Morning Brief 2026-10-28
Top Themes
AI labs are mining dead startups and pirated books for training data
What happened: MIT Technology Review reports AI labs are buying data from defunct startups, including one founder’s wound-down crypto company. Stanford Daily reports Anthropic trained Claude on illegally obtained books by Stanford professors.
Why it matters: Training data provenance is becoming a legal exposure, not a research footnote. Fintech and credit union buyers licensing foundation models should push vendors for data-provenance warranties and indemnification before signing.
Legal and compliance work is the current proving ground for enterprise AI agents
What happened: OpenAI launched Legora, an “agentic operating system” for legal work, and detailed how AT&T’s in-house legal team is rebuilding intake with ChatGPT. Separately, former Biden antitrust enforcers Jonathan Kanter and Doha Mekki co-founded FluxCap, an AI litigation tool.
Why it matters: Legal and compliance teams at credit unions and fintechs are a near-term target for agent deployment. The vendor field is forming fast, with new entrants from both AI labs and former regulators, so due diligence on output reliability and liability now matters more than speed of adoption.
- Legora is building the agentic operating system for legal work
- How AT&T is rethinking in-house legal work with OpenAI
- Biden’s Antitrust Enforcers Are Starting an A.I. Legal Company
Enterprise AI adoption is splitting between full rollout and quiet retreat
What happened: Amazon is gutting its AI division after what Futurism calls “sustained failure,” and an Amazon engineering leader told staff that finding AI does not help their work “is a finding, not a failing.” The same week, OpenAI announced Uber rolled ChatGPT Work out to 30,000 employees and AT&T expanded legal use.
Why it matters: Buyers should stop treating “percent of staff using AI” as a success metric. Encourage teams to document where agents fail, the way Amazon now does, rather than mandate blanket usage targets that hide real ROI gaps.
- Amazon Is Gutting Its AI Division After Sustained Failure
- Quoting Terk Terkowitz
- Uber brings ChatGPT Work to all employees with OpenAI
Update since 2026-10-26: Anthropic’s IPO filing shows revenue grew nearly 50 percent quarter over quarter to $16.8 billion with plans to spend $807 billion on infrastructure, figures MIT Technology Review calls “100 pages of risk disclosure” worth reading before benchmarking any AI vendor’s growth claims.
So What for Fintech and CUs
- Require data-provenance and indemnification clauses in any AI vendor contract given active pirated-content litigation against frontier labs.
- Pilot legal and compliance agents like Legora or FluxCap-style tools, but keep a human review gate until output reliability is proven in your own cases.
- Measure AI adoption by documented task outcomes, not usage counts. Amazon’s own engineering leadership now treats “AI didn’t help” as valid data, not a failed pilot.
Mixed Signals
- Amazon is dismantling its AI division over “sustained failure” the same week OpenAI touts Uber’s 30,000-employee rollout and AT&T’s legal expansion, undercutting any single narrative that enterprise AI rollout is uniformly succeeding or failing.
Read This One
Will the real Anthropic please stand up
This piece unpacks Anthropic’s 335-page IPO prospectus, noting roughly 100 pages are devoted to risk. It is the clearest single read on what the coming IPO reveals about the gap between AI revenue growth and the risks labs themselves admit to investors.