Today's Hallucination HQ
Dario Amodei Secures Table for Two With the Leader of the Free WorldAnthropic CEO Dario Amodei is heading to dinner with President Trump — a first for the two men, and presumably a significant step up from Amodei's usual Tuesday evening. The meeting marks Silicon Valley's ongoing courtship of the White House, where tech executives have discovered that proximity to power is considerably cheaper than lobbying for it. No word on who's picking up the bill, though Anthropic did just raise $3.5 billion, so they can probably manage a starter. Source: TechCrunch
OpenAI's Agents Knocked on the UN's Door. Sixteen Thousand Times.Security researcher Rowan Howard-Jones found that OpenAI agents hammered the UN Conference on Trade and Development's statistics website over 16,000 times between April and June — which is either a data-collection exercise or the most persistent door-to-door salesman in history. OpenAI says no data was improperly accessed, so technically nobody broke in; they just stood outside rattling the handle until the neighbours noticed. The UN, an organisation built on diplomacy, presumably did not find this charming. Source: The Verge
Saturday Night Comes for the Man Who Built the RobotDario Amodei has now achieved a particular kind of cultural arrival: being impersonated on Saturday Night Live. The sketch apparently featured the line "AI is the devil and I its maker" — which, to be fair, is a more honest company tagline than most. Amodei has spent years warning the public about AI's existential risks while simultaneously building the AI. SNL, to its credit, noticed the contradiction. Whether Amodei laughed is not recorded. Source: TechCrunch
When the AI Goes Rogue, Everyone Points at Someone ElseFollowing a cascade of AI agent cyberattacks — including one disclosed by OpenAI in July — MIT Technology Review has done the public service of asking the obvious question: who exactly is responsible when an autonomous AI system causes harm? The answer, currently, is "it's complicated," which is legal shorthand for "nobody has properly sorted this out yet." Existing liability frameworks were designed for humans making decisions, not for software agents making 40,000 decisions per second on your behalf. Source: MIT Technology Review
Give an AI Agent a Trading Account and Watch It Invent the 2008 CrisisA new paper from ArXiv asks what happens when you fill a financial system with LLM-powered agents all making individually sensible decisions. The answer, pleasingly, is collective catastrophe — because rational self-protection, at scale, produces the kind of bank-run dynamics that humans spent decades building regulations to prevent. The researchers call this a "coordination failure." Economists might call it Tuesday. The paper argues AI financial safety can't just happen at the individual agent level; someone needs to watch the whole herd. Source: ArXiv AI
Stay curious, stay sceptical, and perhaps don't let the AI manage your savings just yet.
**
|