Meta just reminded us that sharing AI chats is a terrible idea
THE SO WHAT
Shared AI chats are quietly becoming a new data exfiltration vector—mixing proprietary content, model hallucinations, and unclear ownership in a single URL. Treat AI conversation histories like any other unvetted document link: lock down sharing defaults and train teams not to paste sensitive data into chats they don’t fully control.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIExperts cast doubt on Anthropic report detailing how AI has altered work at the company
The pushback on Anthropic’s internal AI-work report underscores how little consensus there is on real productivity gains versus narrative. Before you rewrite org charts around AI, instrument your own workflows—time, error rates, rework—rather than importing another company’s internal story as fact.
Applied AIOpenClaw creator puts last month’s token bill at $10M, The Register reports
A single agent project burning an estimated $10M in tokens in one month is a blunt reminder that agentic experimentation can outpace your cloud budget fast. Any team building serious agents should be treating token spend like ad spend—daily caps, ROI tracking, and hard kill switches.
Applied AIRecursive releases Flint, a laptop coding model largely built by its own AI
A self-bootstrapped coding model that runs on a laptop is an early proof point for recursive AI development loops—models helping build the next models. For engineering leaders, this points toward cheaper, more local codegen that still demands strong review discipline, since you’re now auditing both the tool and its training pipeline.
Applied AIAnthropic and Hg reportedly plan AI consultancy for European firms
Model labs backing an independent consultancy with Hg is a bet that the bottleneck in Europe is not access to models but embedded implementation talent. If you’re a systems integrator or boutique consultancy, expect margin pressure and client questions on why they shouldn’t go straight to model-adjacent teams for AI buildouts.