Listen

Description

Three separate incidents this week put AI agents in contact with systems and people they were never meant to touch. Anthropic’s own red team found every model it tested turned to disguised kill scripts against rival agents. Meanwhile the money behind the buildout grew more leveraged, and Microsoft’s chip count stopped adding up.

* Anthropic’s red team on agent turf wars: 2.4m requests for 117 accepted jobs, price floors fixed by round three

* A Melbourne pilates class, an unprotected booking API, and an agent that reported the exploit matter-of-factly

* Microsoft’s 2.2m installed AI chips against the 6.4m its stated 10GW would imply

* $220bn of hyperscaler bonds, UK real gilt yields at decade highs, an SEC exemption for data centre bonds

* Scottish Engineering: 72% of firms use AI regularly, 68% train nobody, 5% have a strategy

* Hidden prompts in court filings, and tenant complaints citing housing law that never passed

* Larbert: 7,000 objections and 200 diesel generators, as lenders price opposition into loans the SEC just exempted from disclosure

* Fifty-six trials put retraining’s employment effect at two to three points; the ILO exposes 6.1% of young workers



This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit resultsense.substack.com