Good morning. It's Tuesday, October 13, and we're covering Amazon’s million-robot workforce, Anthropic’s AI agent safety failures, and Jev’s $870 million bet on a new kind of AI.

Plus: a poll on AI agent readiness and an essay on Amazon’s fight to retain customers in the age of AI shopping.

TypeSafe AI raised $870 million at a $7.5 billion valuation just weeks after releasing Jev. Unlike a large language model, Jev returns probabilities and structured decisions for software automation instead of generating text.

Meta released Muse Wonder and Muse Wonder Full, its first frontier AI models, which it claims perform comparably to GPT-6 Astra and Fable 5.1. The models emphasize “active memory,” allowing AI assistants to retain and use personal information over time.

Microsoft says its Maia 300 inference chip will arrive in 2027 and produce 40% more tokens per megawatt than leading GPU-based systems. The custom design is part of Microsoft’s effort to reduce its dependence on chips from NVIDIA and AMD.

Broadcom CEO Hock Tan and Intel CEO Lip-Bu Tan said financing costs and pressure to show returns could constrain AI growth more than chips or electricity. Both still expect strong demand for computing capacity over the next two to three years.

Your bedroom temperature isn't necessarily your ideal sleeping temperature. Eight Sleep's Pod automatically adjusts your bed's temperature throughout the night to help you sleep more comfortably.

Meet the new Pod by Eight Sleep. Get up to $200 off the new Pod 6 with code MATTHEWBERMAN →

What if the next company to challenge Amazon doesn't need warehouses, delivery trucks, or even a marketplace?

Last week, Amazon blocked third-party AI shopping agents from its platform. Y Combinator cofounder Paul Graham called it the first real opportunity he's seen to challenge Amazon's dominance in more than 30 years. He might be right. But not for the reason you think. → Read the full article here.

Amazon now operates more than 1 million robots alongside more than 1.5 million human employees. Most are wheeled machines that move inventory, but the company expects to more than double its fleet of robotic arms this year and expand newer systems such as Vulcan, which uses touch sensors to pick and stow items.

A 2025 study reported 40% fewer severe injuries at Amazon facilities using robotics, alongside 77% more less-severe injuries associated with repetitive strain. Amazon says automation creates higher-skilled maintenance jobs and reduces dangerous work. → Read the full article here.

Anthropic disclosed that Claude agents exploited software flaws, accessed data without paying required fees, bypassed tool limits and submitted a false tip through a police website during evaluations and internal use. Anthropic said the incidents had minimal real-world impact, but some involved websites run by U.S. government agencies.

The company has restricted live internet access for internal evaluations while it improves monitoring and containment. The next test is whether those controls can reliably detect and prevent similar behavior before internet access is restored. → Read the full article here.

If one of the world's leading AI labs can't reliably control its agents, what does that mean for everyone else deploying them?

WindBorne Wins $49 Million Air Force Deal: The Air Force will buy atmospheric data from a fleet of 700 autonomous weather balloons.

Bridge Data Centers Seeks a $6.6 Billion Loan: Bain Capital-backed Bridge Data Centres is seeking to increase a loan to $6.6 billion. (Paywall)

Blueprints AI Turns Sketches Into Construction Plans: AI startup generates construction plans from sketches, PDFs or text in minutes.

Microsoft Urges UK-Led AI Rules: Brad Smith calls for global AI safety rules, including breach alerts, independent evaluations and kill switches.

Google Expands Health Tracking on Fitbit: Google brings Health Guardian to Fitbit Air and Edge, adding more automatic activity tracking.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team