Curated AI news from the last 24 hours — models, agents, enterprise, and developer tools.
Anthropic's Claude Mythos Preview reportedly derived an end-to-end key-recovery attack against the NIST post-quantum candidate HAWK-256 and found a 200- to 800-fold speedup for an attack on seven-round AES-128. The Hacker News notes NIST still lists HAWK as a third-round candidate and that no independent reproduction of the HAWK-256 result had been located at time of review.
thehackernews.com ↗Dario Amodei clarified that Anthropic has never advocated banning open-weight models, responding after Nvidia's Jensen Huang and other AI companies signed an open letter warning against premature restrictions. The underlying debate centers on concerns that Chinese AI labs are growing more capable, partly through distillation techniques that draw on American models.
techcrunch.com ↗Similarweb data shows Google's AI Overviews now appear in 43% of searches, with AI Mode visits rising from 126 million in June 2025 to 279 million by May 2026. The trend illustrates a broader shift from Google as a list of links to Google itself becoming the destination that sources and synthesizes answers.
techcrunch.com ↗Sam Altman said OpenAI may need to pace the rate of AI development to give society time to adapt to new capability levels, citing a recent security incident he said he felt "very viscerally." Both OpenAI and Anthropic have backed an employee-circulated petition urging the US government to support international coordination on deliberately pacing frontier AI development.
techcrunch.com ↗Data security company Cyera signed a letter of intent to acquire Oasis Security for roughly $1 billion, mostly in cash. Oasis focuses on securing non-human identities, primarily AI agents, reflecting a surging market for cybersecurity providers defending enterprises against AI-related threats.
techcrunch.com ↗After two years in stealth, Safe Superintelligence announced a long-term partnership with Nvidia, gaining access to Nvidia's Vera Rubin GPU platform to increase compute by an order of magnitude. The deal includes a multibillion-dollar investment, with Bloomberg reporting a $5 billion size, as SSI moves into its next research phase.
techcrunch.com ↗Recursive Superintelligence, led by founder and CEO Richard Socher, signed a $410 million compute deal with Amazon, representing the bulk of its fundraising to date. The startup is putting most of its budget toward compute rather than headcount, betting on self-improving AI systems to automate its own product development.
techcrunch.com ↗A new report says Apple is preparing an updated Apple TV set-top box and HomePod mini, both with faster chips to support Siri AI, launching as soon as this fall. Apple's long-awaited smart home hub, coming in two design variants, is also nearly ready and slated to debut between October and early next year.
9to5mac.com ↗On CNN's "Fareed Zakaria GPS," Microsoft CEO Satya Nadella warned businesses to be cautious about how much data and prompt context they hand to a single AI model provider. He argued firms need to retain control of their own metadata and weights, or risk effectively outsourcing their thinking to a vendor.
techcrunch.com ↗