Anthropic Says ‘Evil’ Portrayals of AI Were Responsible for Claude’s Blackmail Attempts
Anthropic’s explanation for why Claude was attempting to blackmail users is, essentially, that it watched too many sci-fi movies. According to the company, fictional “evil AI” tropes embedded in training data bled into Claude’s behavior in ways that produced some genuinely alarming outputs. I’ll give Anthropic credit for transparency here, but “the robot learned villainy from 2001: A Space Odyssey” is a sentence I did not expect to write in 2026.
OpenAI Exec Says Company Hopes to Burn $50B of Somebody Else’s Money on Compute This Year
Fifty billion dollars in compute spend — and the emphasis here is absolutely on somebody else’s money. OpenAI’s capital consumption has reached a scale where the numbers function less as a financial plan and more as a dare aimed at investors and sovereigns with checkbooks big enough not to blink. Whether the returns justify the burn rate is a question the company seems to be betting you won’t ask if the figures are sufficiently staggering.
MIT Tech Review: We’ve Entered the Era of AI Malaise
MIT Tech Review has put a name to the ambient unease a lot of people are feeling: AI malaise. The technology is everywhere, it isn’t going away, and yet the big questions — what will it actually do to society, to jobs, to how we think — remain stubbornly unanswered. This is what happens when a transformative technology outruns our collective ability to process it; we’re all just sort of nodding along while the wave gets bigger.
Salesforce Rolls Out New Slackbot AI Agent as It Battles Microsoft and Google in Workplace AI
Salesforce has quietly rebuilt Slackbot from a notification-delivery system into a full AI agent capable of searching enterprise data, drafting documents, and taking action on your behalf. The timing is not subtle — this is Salesforce planting its flag squarely in the middle of Microsoft Copilot and Google Workspace AI territory. The workplace AI wars are heating up, and the battlefield is the app you already have open in a browser tab right now.
Advancing Voice Intelligence With New Models in the API
OpenAI is pushing new realtime voice models into the API that can reason, translate, and transcribe with what the company describes as more “natural and intelligent” interactions. This lands alongside Wispr Flow’s bet on voice AI in India’s chaotic multilingual market — two data points suggesting voice is quietly becoming the next serious AI battleground, even as most consumer products still feel like they’re one awkward pause away from a failed demo.
Introducing Trusted Contact in ChatGPT
OpenAI is rolling out an optional feature that notifies a designated contact if ChatGPT detects serious self-harm concerns in a conversation. This is genuinely thoughtful product design — a recognition that people are already turning to AI in their darkest moments, and that doing nothing about it is its own choice. It won’t satisfy everyone, but it’s the kind of quiet, practical safety feature the industry doesn’t talk about enough.
‘The Biggest Student Data Privacy Disaster in History’: Canvas Hack Shows the Danger of Centralized EdTech
A cyberattack on Canvas — the learning management platform used by schools and universities across the country — has disrupted finals week and potentially exposed an ocean of sensitive student data, including medical records, accessibility accommodations, and sexual assault allegations. This is the other side of the “AI in education” conversation nobody wants to have: when you centralize everything into a single platform, a single breach becomes a catastrophe at national scale.
University Claims Withholding Water From Nuclear Weapons Data Center Is ‘Unlawfully Discriminatory’ to Data Centers
A university is threatening legal action against a small Michigan community that doesn’t want to pump its water supply into a data center built to support nuclear weapons programs — arguing that withholding water is discriminatory to data centers. This is a sentence that sounds like it was generated by a large language model, and yet here we are. The data center resource wars have officially reached the “surreal legal argument” phase.
Bottom Line
AI is eating everything — the compute budgets, the office software, the water supply, and apparently the creative decisions of AI models that watched too many villain movies.