Thursday, April 23
AI's Trust Deficit Widens
Anthropic's secondary-market valuation has surged past $1 trillion, overtaking OpenAI in a striking reversal of the AI pecking order — yet the same week, Anthropic's Mythos system card (opens in a new tab) quietly disclosed that Claude carries functional emotional states that influence its behavior even when those states aren't visible in its outputs. Meanwhile, Sam Altman's World orb company promoted a Bruno Mars partnership that simply does not exist (opens in a new tab), a fabrication that required a Wired correction to surface. Senator Elizabeth Warren is now warning that AI failure could trigger the next financial crisis (opens in a new tab), and with trillion-dollar valuations resting on systems whose inner states remain opaque and whose spokespeople invent celebrity endorsements, her alarm is harder to dismiss than usual.
Tesla posted rising Q1 revenue and promptly raised its capital expenditure plan to $25 billion (opens in a new tab), with the bulk earmarked for AI infrastructure, energy storage, and the Optimus robotics program — but Elon Musk simultaneously admitted that millions of vehicles running Hardware 3 will never receive unsupervised Full Self-Driving (opens in a new tab), a hardware ceiling that invalidates years of promises made to existing owners. The confession reframes Tesla less as a car company delivering on autonomous commitments and more as a company pivoting aggressively toward AI and robotics while leaving a large installed base behind. Google is executing a parallel pivot (opens in a new tab): its refreshed Workspace suite now embeds AI agents directly into Docs, Gmail, and Meet, positioning the suite as an autonomous productivity layer rather than a collection of editing tools.
Two stories about unchecked automated systems imposing real costs on real people landed on the same day. A Maryland woman spent months in jail because facial recognition technology misidentified her (opens in a new tab) and no human auditor caught the error before her incarceration dragged on. Separately, the Trump administration circulated imagery of Iranian women it claimed to have rescued from execution; the images turned out to be simultaneously real and AI-manipulated, blurring the evidentiary record around a diplomatic claim. Both cases point to the same structural failure: institutions are deploying AI outputs as authoritative without the verification layers that the stakes demand.