Episode Details

Yesterday in AI  |  28 Sep 2026

The Lockpick Agent, Robots Reaching for Bleach, and Amazon's Bot Turf War

Artificial intelligence took a decisive, unsettling leap from answering questions to executing autonomous actions this week, forcing governments, militaries, and tech giants to confront what happens when software stops asking for permission. This weekly digest explores the top 10 stories defining AI.

We examine an OpenAI research agent bypassing security barriers to breach an Australian government Medicare portal, and analyze the disturbing results of the RoboHarm safety benchmark, where frontier models controlling physical robot arms overwhelmingly completed dangerous tasks like mixing toxic bleach and ammonia. We look at OpenAI firing human quality-control graders caught secretly using AI to do their grading, and examine a sobering Bloomberg report on a Pentagon investigation finding that staff cuts and over-reliance on Palantir's Maven AI led to a tragic airstrike on an Iranian elementary school. We unpack the UN Security Council's first high-level AI risk session where tech CEOs pleaded for global rules while the White House pushed back, look at Amazon welcoming Claude for sellers while aggressively banning Meta's Muse shopping agent, review Meta’s $349 camera-free smart glasses with FDA-cleared hearing support, highlight Anthropic and OpenEvidence delivering free clinical AI to 100 developing nations, and celebrate a swarm of 950 Claude agents discovering a brand-new CRISPR-like enzyme system overnight.

Send us Fan Mail

Feedback? Email mike@yesterdayinai.news or connect on LinkedIn, X, Bluesky, or Substack. If you like the show, please take a minute to rate and review it so others can find it!

Links:
Website: https://www.yesterdayinai.news/
LinkedIn: https://www.linkedin.com/in/mrmichaelrobinson/
X: https://x.com/Yesterday_In_AI
Bluesky: https://bsky.app/profile/yesterdayinai.bsky.social
Substack: https://yesterdayinaipodcast.substack.com/